You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@mxnet.apache.org by jx...@apache.org on 2018/01/19 00:35:07 UTC
[incubator-mxnet] branch vision updated (7c92b98 -> 8eec536)
This is an automated email from the ASF dual-hosted git repository.
jxie pushed a change to branch vision
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git.
discard 7c92b98 [Vision] add test cases for flip, normalize, to_tensor (#8919)
omit 8657ad0 add comments and sanity check (#8901)
omit 2205b22 fix (#8857)
omit 06b4f47 Vision (#8856)
omit 9cebe62 [WIP]hue (#8678)
omit 86f8799 image flip op (#8759)
omit f33654b [Image] add random lighting (#8779)
omit 4765139 Refactor image operators (#8761)
omit e9ae848 Revert "add test script"
omit 23f6827 add test script
omit b8b0d71 fix (#8736)
omit db2f1f4 [Image OP] Normalize (#8731)
omit 303e882 image to_tensor (#8691)
omit e741d8f Add Gluon data transform (#8672)
omit ae9fd42 fix image_random compile (#8665)
omit 2c7ceba [WIP][vision]fix (#8657)
omit 5d3460d add stub (#8650)
omit d1a569c [WIP]]Vision (#8649)
omit f50e7eb [WIP]Image Augmenter (#8633)
add 4dc8933 Fix docstring format (#8734)
add 1f63a9a Revert "Support sparse for custom python operators (#8620)" (#8733)
add cd47962 2bit gradient compression (#8728)
add 068b589 Kernel operator tuning (#8686)
add cace29f NCCL integration (#8294)
add bbe078e [cpp-package]Update readme#8655 (#8746)
add 5c328f1 comment typo: there are there (#8753)
add 12c3fb1 fix lint with cmake (#8752)
add 1852e2f Add cast to Block and Parameter. Implicit dtype casting is removed. (#8735)
add 9d48f74 add code signing key (#8743)
add eb72d72 Added a security best practices doc (#8744)
add ec6144f a user friendly way to use g2c in module and an example of g2c (#8632)
add 8df20a2 [scala] EvalMetric sumMetric is now a Double instead of a Float (#8297)
add abf01b4 Remove spureous std::move, fix warning regarding RVO being prevented (#8755)
add 5518986 Explicitly convert float value (#8758)
add 89d2a71 Updating ps-lite submodule (#8769)
add e835777 Fix doc link (#8776)
add 5ad6eab [CMAKE] Fix cmake library path when installing python package (#8680)
add 1f6d586 [ImageIO] Fix image io for opencv3.3 (#8757)
add be18794 Make make lint compatible with python3 (don't call python2 explicitly) (#8401)
add a0031ac cast scalar value in invoke to float (#8778)
add 1264313 Add Intel openmp as a submodule and build for x86 architectures (#8730)
add 3953e79 Fix weird hang bug due to cuInit sometimes calls fork (#8790)
add 0bf76a8 Fix weird hang bug due to cuInit sometimes calls fork (#8804)
add 5088477 remove meaningless type qualifier (#8774)
add 3c32f76 fix build (#8788)
add 65b57b1 Fix a typo (#8809)
add 2ccc6b6 fix a typo (#8811)
add 8bd65fd Fix warnings (#8813)
add fdc0766 multi-device support (#8812)
add 05047ad Tune without Launch specialization macros (#8719)
add c462d34 Python 3 compatiblity during optimizer serialization. (#8334)
add 2353229 fix building warnings below (#8820)
add 91ffd69 Update test_operator_gpu.py (#8819)
add 3226bce mxent link fix (#8824)
add ce4b921 fix https://github.com/apache/incubator-mxnet/issues/8262 (#8838)
add 7aea2ef fix search (#8841)
add 632c897 Refreshed branch fix_omp_link (#8842)
add 2f8c1e8 add inline for cached op & fixed a bug when calling backward on variable (#8701)
add 5b00a5f change default arc to Common (#8624)
add 47a71a6 Profiler: set cpu/gpu num during execution (#8722)
add 0df7953 Moved cub to 3rdparty (#8878)
add 6d6c4ea fixed broken links on master (#8879)
add 89c4db7 Use https for Jquery file (#8855)
add cabbcff Fix typo. (#8888)
add 2aa506a Move legacy operators under the nn directory (#8858)
add 319494f Change the initialization for default constructor to avoid using explicit constructor (#8854)
add ae0fe77 fix reshape check (#8893)
add d7da05b fix race when temp space is used in copy & fix instance overwrite in g2c (#8867)
add f13ce52 doc improvements for default param 'TShape()' (#8906)
add 3b3016e Caffe to MXNet code translator (#8782)
add 1dc4aea fix multiprocessing too many open files (#8904)
add ddb4782 Update index.md (#8760)
add 7639e2b fix documentation homepage typo (#8931)
add 2a85cd4 Add myself as contributor. (#8929)
add 14cd740 Remove finalizers from Scala API (#8887)
add 95a226d Fix unit test build (#8946)
add 6c51629 rsp[:] = const_value, used in constant initializer for rsp weight (#8891)
add 6f403c0 Initial Prep for 1.0: bump up version and add 0.12.1 changes to master (#8704)
add 5f8efb0 NDArray Indexing tutorial and Gradient Compression FAQ (#8766)
add b909769 [R] various small changes for efficiency and robustness in R package (#8118)
add 43d534d Small first iteration for profiler: use nonblocking queue (#8803)
add 7484437 [Do not merge] Move to new CI (#8960)
add c965c8b Change type of 'lr_refactor_ratio' to 'float' or else the learning-rate-refactor will never work (#8982)
add 5c11ecd Fix race condition in engine start/stop (#8995)
add c94b104 Implement assignment operator on Timing 'info' struct (#8996)
add f154101 Some fixes for example/reinforcement-learning/parallel_actor_critic (#8991)
add b652a20 Fix multi task (#8987)
add 075ef68 Fix 'from_layers' of resnet101 by changing names of extract-layer from base network (#8998)
add c9e440d Reduce the verbose warnings in CI (#8981)
add 24ea5b0 Factorization machine example & sparse example folder re-org (#8767)
add 9d4bb9c Minor changes to Caffe Translator (#8939)
add d77d937 Fix __repr__ for gluon.Parameter (#8956)
add 918fac3 Update PULL_REQUEST_TEMPLATE.md (#9003)
add 3d34f2e [CI] Fix indentation of Jenkinsfile, no code changes (#8988)
add e692d58 Updates for R readme and example (#8984)
add 7d610f8 Correct the alexnet url. (#8975)
add 01b7ae3 change workspace allocation in random samplers (#8935)
add 1c22bc7 [clang][warning] fix warnings about braced initializer (fixes warning in concurrenc… (#8870)
add 84fde4d Locate liblapack.a in /usr/lib64 (#8073)
add ac4369c update readme (#9005)
add 3ee1ca2 Fix Example example/dec/dec.py (#8990)
add 8a8e770 Pylint fixes (#9013)
add 910c835 Delete recommendation_systems.md (#9009)
add 20666c3 Add wide and deep model into sparse example (#8180)
add 2700ddb Symbol __getitem__ using list_outputs() is too expensive (#8989)
add 5284513 Update for MXNet 1.0. PEP8 fixes to code and misc improvements. Tested under Python 2.7 and Python 3.6 on Sagemaker. (#9036)
add 06ed4dc Add truncated bptt RNN example using symbol & module API (#9038)
add 3901f5f Use backtrace from dmlc-core with demangled C++ symbols on segfault (#8851)
add db15049 Remove redundant or unused imports (#9021)
add 167871a fix small memory leak of sparse embedding (#9025)
add 087a906 Adding .png to the list of allowed extensions in the im2rec.py (#9043)
add ca79cfc Fix Gan tutorials with symbolic interface (#9030)
add 8f79651 update autograd.Function docstring for usage (#9039)
add 6d1f07d Remove defunct NLP tutorial page (#9045)
add 9e2bdeb Disable test_optimizers:test_sgd (Temp fix #9000) (#9001)
add 73fa65f Add cscope filenames to gitignore (#9019)
add 85a5b8e fix VGG error in benchmark_score (#8664)
add 41394a4 Disable test_operator_gpu.test_svmoutput_with_type (#9059)
add fdd41a8 Add googletest as a 3rdparty library (#9016)
add d43f4f9 improve image io example (#9061)
add 1f70a48 Updating the build documentation with instructions to build mxnet with NCCL (#9058)
add c4a76da Implement Khatri-Rao operator (#7781)
add 49c60fd [example] fix neural-style end2end example (#9063)
add bfa0989 Fix examples of profiler and cpp-package (#9065)
add 8623bab add CapsNet example (#8787)
add 392e7ac Increase git timeout for CI (fixes #9070) (#9071)
add a8862a7 Remove torch support (#9072)
add 09ed385 square sum gpu impl (#9014)
add 1c8e066 Restore submodules to dmlc (#9069)
add 48f62cf Fix Khatri-Rao test build (#9081)
add 5540ee9 Fix check against nonexistent variable (#9082)
add 24c7231 Modify Caffe example to use module interface instead of the deprecated model interface. (#9095)
add 2a4505a add hyper parameters alpha and resolution, add extra compact symbol with less lines of code (#9099)
add 3b7134a Some changes to numpy_ops examples - (#9092)
add d23b91d added assertion to validate accuracy (#9107)
add 5858d62 Remove defunct demo (#9060)
add df25378 add shared storage in windows (#8967)
add fb254c4 Updating the Python readme (#9075)
add b53a13e FCN example updates (#9066)
add 5dcdd5d Update linear-regression.md (#9103)
add c6d4e7f image-classification example fix python3 compatibility (#9053)
add b9b4532 More details to the windows build process (#8519)
add 9b0a5ca csr slice operator, gpu implementation (#8814)
add 73dba33 [BugFix][CoreML Converter] Dense layers w/o bias. (#8684)
add df55829 Fix compatibility of examples with py3 (#8993)
add 16bd961 Usability fixes for examples (#9091)
add d1d3e97 Fix example/reinforcement-learning/dqn (#9128)
add 500ed5f fix multi worker dataloader deadlock (#9126)
add 9272d9c Fix nadam (#9127)
add ae70769 add zero-grad for rounding ops (#9040)
add a38ab6b Install docs - default to install CUDA 9 and cuDNN 7 (#9147)
add b59943e warning autotune show once (#9132)
add 521825d Fixed incorrect wording in mnist tutorial (#9125)
add c71a2f3 Usability improvement bi lstm sort (#8944)
add 080d29c Fix float16 min and max (#9149)
add 5c3acff Add Wikitext for gluon/rnnlm (#9090)
add ddec3cc R RNN API fixes and Optimizer clip gradient on NDArray (#9022)
add c6cdf51 Merging version changes from 1.0.0 to master (#9168)
add b91ccdc example/nce-loss fixes and improvements (#9093)
add f8af6f5 LicenseUpdates from 1.0.0 to master (#9170)
add 710d0f6 updated master branch with 1.0.0 release on the home page (#9173)
add bccf9fa Remove duplicate material (#9080)
add d90c1b7 bump up version to 1.0.1 (#9183)
add ef85063 Dropped Sergey's explicit copyright note per https://lists.apache.org/thread.html/3cc041b688e9efd235ebfad8336917b7e1566b4acceff7783cf09c3c@%3Cdev.mxnet.apache.org%3E (#9191)
add 3219620 Mkl fix pr (fix an issue in prepare_mkl.sh and add test case) (#9153)
add 17427d1 Make versions of python dependencies like numpy, deterministic. (#9169)
add ad1ed7c fix incude error in plugin/opencv/cv_api.cc (#9206)
add 911c81e Support negative axis in concat (#9204)
add 01a2a40 fix parameter overwrite in _generate_symbol_function_code (#9180)
add 02b1131 Rnn example updates (#9074)
add f20a205 Install docs - update info about DLAMI (#9150)
add be40d31 refactor graph exec (#9122)
add 7f1dd77 Usability improvement for cnn_chinese_text_classification (#9027)
add 2586c66 Modified nce-loss README to use image files in dmlc/web-data repo (#9212)
add 34a5195 fix random generator: do not gen seed each time (#9119)
add 35781a4 Add result sections in sparse tutorials (#9085)
add d0388d0 Update mkl (#9218)
add f11e591 fix autograd import path (#9221)
add 4c15f58 fix link error for parallel rng (#9256)
add 7e1c227 fix bayesian-methods example (#8958)
add e3e1f23 raise err in io.NDArrayIter for invalid usecase (#9228)
add e4128ec Fix website parsing error (#9260)
add 25dd12d Adding multiple pretrained models (#9215)
add 79cb705 Lint enhancements to R demo scripts (#9270)
add 3f10bd4 fix typo for assert messages in gluon/nn/conv_layers.py (#9277)
add 26e81d6 Add documentation generation to PRs (#9281)
add 19f8e2b Add storage fallback msg to copyfromto (#9219)
add 57d1692 Modify R package gitignores (#9269)
add 4aff838 Update deadsnakes ppa name (#9268)
add 5b99b25 1 d conv with cudnn (#9184)
add 5ab629f Add timeouts to windows tasks in Jenkinsfile (#9280)
add dc48eb5 Update module.md to include results of the code snippet (#9123)
add ef82fd2 Update broadcast_reduce_op_value.cc (#9287)
add 4e3ac33 update front-page model zoo to gluon model zoo (#9286)
add 12cb0d2 FTML optimizer implementation (#9262)
add 728e5c2 fix trainer: init kvstore before load states (#9224)
add 7e7cf3a Contrib operators for object-detection bounding box related stuffs (#8603)
add 996d548 update README.md with details on typical usage of the scripts (#9108)
add 600a163 Updating the README files and examples in "ctc" and "recommenders" folder. (#9073)
add 9eff221 Usability improvement mxnet vae (#9008)
add 9d51e1a sparse output for binary scalar op with zero (#9227)
add f79f0d4 Make numpy functions return Symbol instead of numpy object array (#9293)
add 8505442 Add operator for dot(dns, csr) = csr (#8938)
add 546b73a Add clarification in sparse api doc (#9282)
add 2df29cc Updating the PyPi package URL to be Apache and not DMLC (#9299)
add 661bcef bugfix for parallel rand generator on multi-gpu (#9300)
add 469955d Increase timeout for GIT on windows (Fix #9297) (#9298)
add 5e587e5 fix w&d (#9303)
add 4988535 add epoch parameter to export() of HybridBlock (#9290)
add c3616ed Add label_from_zero_one argument to LogisticLoss (#9265)
add d918868 Fix the gradient of gather_nd (#9200)
add 1602b12 Add Marco de Abreu to CODEOWNERS related to CI (#9309)
add 9be50e0 Fix CODEOWNERS syntax (#9313)
add df9f79a standard update for sparse sgd_mom_update (#9189)
add f9fd88b fix utf8 error in odc (#9322)
add b596cf6 Prepend LD_LIBRARY_PATH to dll_path in find_lib_path (#9311)
add 004dead Fix custom op multi-gpu scaling (#9283)
add 898eab3 added assert for achieved accuracy (#9106)
add 2d28593 example/autoencoder fixes for MXNet 1.0.0 and pylint and addition of README (#9097)
add acff519 updated comments for mnist tutorial (#9104)
add c3d6cf0 Fix example/module folder and remove duplicate examples (#8964)
add deb25bc Refactor scala code/tests/examples/docs to use org.apache.mxnet (#9324)
add 3463967 change username javelinjs -> yzhliu so as to be more easily mentioned (#9351)
add 8ad77f3 Temporarily disable test with failing http connection (#9333)
add 8168292 fix hard-coded in_channels in gluon model zoo resnet (#9341)
add 9543db4 fix path separator (#9352)
add 243ea39 Fix CUDA auto-detect and cxx flags (#9362)
add 50bfdc7 added updated examples to list and contrib info (#9356)
add 492a5fd add MSRAPrelu initializer for cpp-package (#9121)
add f30bb4a Use provided scope for log4j dependencies (#9365)
add 2d49d59 Fix warnings on cpp-package (#9089)
add 46cf856 Fix unit test build (#9371)
add fd5e5fe Update dependencies / Makefile + add Maven profile to support build with Scala 2.12 (#9339)
add a80245d improve memory usage in sparse.dot (#8826)
add 6c1f4f7 Add mxnet.text APIs (#8763)
add 2b73aac Add clang jobs to the CI (#9330)
add 3eae3bb Update MKLML to v0.12 (#9383)
add d511d98 Remove using namespace mshadow from header file (bad practice) (#9321)
add 1732460 Fix C++ unit tests (#9349)
add 51ea815 Revert "Temporarily disable test with failing http connection (#9333)" (#9379)
add 4600070 support regex of collect_params() (#9348)
add 5106985 example/ctc improvements (#9345)
add 68750bc Update test_kvstore.py (#9386)
add 5251b86 Signum optimizer (#9220)
add 67cfbc8 Revert "Add mxnet.text APIs (#8763)" (#9401)
add f129992 adding missing modules to DEC (#9407)
add 47a3422 Update operator.py (#9402)
add e4957c3 resubmit #8763 #9394 to contrib (#9406)
add b1ad161 Fixed memory leak (#9400)
add 171d717 bitrot fix. (#9414)
add 786e376 rsp push and rsp pull for comm device, used in kvstore('device') (#8732)
add 134b205 Use a better formula to calculate sigmoid_bce and logistic_loss (#9404)
add 6dcb0eb move concurrent/identity blocks to contrib (#9427)
add b623488 Fix warnings on unused variables (#9432)
add dea19e3 Add recommendation to use RAII to the contributors guide (#9439)
add 3454974 Add discuss link for asking questions to the README (#9437)
add 1b5caf6 Fix undefined behaviour in memory_test.cc, use of delete instead of delete[] to free arrays through unique_ptr. (#9435)
add 63394de header updates to clarify use of CODEOWNERS (#9426)
add 52f5330 Update index.md (#9424)
add a4a09e5 Random seed setting (#9409)
add 9cc8ea3 Re-enable all op segments when in batch mode (#9055)
add 3ac5376 Batching improvements for GEMM/TRSM operators and full MKL usage docs. (#8846)
add e36a945 raise warning when we detect Block inside nested list/dict (#9148)
add 262c74c Introduce the ENABLE_CUDA_RTC build option (#9428)
add 23f973b amalgamation library fixes for python 2.7 and 3.6 compatibility (#9346)
add 25c7232 Update install-mxnet-osx-python.sh (#8825)
add 11cb609 removes python path insert of tests folder for examples (#9151)
add 894c6de Revert "header updates to clarify use of CODEOWNERS (#9426)" (#9442)
add 02b69db Limit the test_nccl.py to run on 8 GPUs (#9441)
add c74cf1b enabling multithreading in broadcast_reduce (#9444)
add 8054290 Fix crash when opening an image, fix exception safety. (#9370)
add 59700b0 np.fromstring: DeprecationWarning (#9445)
add d35c3d8 Test backtrace demangling functionality (#9434)
add 16746c1 Refactor dropout operator to use ParallelRandom generator and also react deterministically when seeding (#9366)
add f960522 Sequence Operator Improvements (#9306)
add b396bc3 text: use _contan, import modules, and keep code style consistent (#9446)
add 89b4fa0 correct usage of bool arguments from command line (#8880)
add a182aec Update model_parallel_lstm.md (#9447)
add 9353c2e Remove empty lines at the end of file (#9459)
add b023bd5 Add CMake CUDA build job to the CI. (#9329)
add 46c9f48 Rename kvstore/utils.* to kvstore/kvstore_utils.* (#9463)
add a937d98 [CMake] Add im2rec executable (#9456)
add 9ae63d1 standard adam update for sparse tensor (#9468)
add 49846f0 Glossary takes token_indexer and token_embedding in its constructor (#9471)
add 55ce97c added mac gpu install; refactored windows install (#9353)
new fa7575a [WIP]Image Augmenter (#8633)
new 54fa028 [WIP]]Vision (#8649)
new c247df5 add stub (#8650)
new 627ba18 [WIP][vision]fix (#8657)
new 383f490 fix image_random compile (#8665)
new 80f323a Add Gluon data transform (#8672)
new d5532bc image to_tensor (#8691)
new a399908 [Image OP] Normalize (#8731)
new 7996da9 fix (#8736)
new 1873367 add test script
new 400c380 Revert "add test script"
new 05a5974 Refactor image operators (#8761)
new 3c99773 [Image] add random lighting (#8779)
new e97ef36 image flip op (#8759)
new 9e36e30 [WIP]hue (#8678)
new d07f3af Vision (#8856)
new 9a7e3e6 fix (#8857)
new fc940fb add comments and sanity check (#8901)
new 8eec536 [Vision] add test cases for flip, normalize, to_tensor (#8919)
This update added new revisions after undoing existing revisions.
That is to say, some revisions that were in the old version of the
branch are not in the new version. This situation occurs
when a user --force pushes a change and generates a repository
containing something like this:
* -- * -- B -- O -- O -- O (7c92b98)
\
N -- N -- N refs/heads/vision (8eec536)
You should already have received notification emails for all of the O
revisions, and so the following emails describe only the N revisions
from the common base, B.
Any revisions marked "omit" are not gone; other references still
refer to them. Any revisions marked "discard" are gone forever.
The 19 revisions listed above as "new" are entirely new to this
repository and will be described in separate emails. The revisions
listed as "add" were already present in the repository and have only
been added to this reference.
Summary of changes:
.github/PULL_REQUEST_TEMPLATE.md | 12 +-
.gitignore | 4 +
.gitmodules | 8 +-
cub => 3rdparty/cub | 0
3rdparty/googletest | 1 +
3rdparty/openmp | 1 +
CMakeLists.txt | 122 +-
CODEOWNERS | 13 +-
CONTRIBUTORS.md | 10 +-
Jenkinsfile | 861 +-
KEYS | 59 +
LICENSE | 30 +
MKL_README.md | 19 +
Makefile | 51 +-
NEWS.md | 65 +
NOTICE | 8 -
R-package/.gitignore | 5 +-
R-package/DESCRIPTION | 2 +-
R-package/R/callback.R | 22 +-
R-package/R/initializer.R | 21 +-
R-package/R/lr_scheduler.R | 16 +-
R-package/R/metric.R | 2 +-
R-package/R/mlp.R | 16 +-
R-package/R/model.R | 86 +-
R-package/R/model.rnn.R | 46 +-
R-package/R/mx.io.bucket.iter.R | 34 +-
R-package/R/optimizer.R | 58 +-
R-package/R/rnn.graph.R | 379 +-
R-package/R/rnn.infer.R | 117 +-
R-package/R/util.R | 2 +-
R-package/R/viz.graph.R | 24 +-
R-package/README.md | 4 +-
R-package/demo/basic_bench.R | 19 +-
R-package/demo/basic_executor.R | 25 +-
R-package/demo/basic_kvstore.R | 12 +-
R-package/demo/basic_model.R | 56 +-
R-package/demo/basic_ndarray.R | 29 +-
R-package/demo/basic_random.R | 2 +-
R-package/demo/basic_symbol.R | 14 +-
README.md | 5 +-
amalgamation/amalgamation.py | 129 +-
amalgamation/mxnet_predict0.cc | 16 +-
amalgamation/python/mxnet_predict.py | 5 +
benchmark/python/sparse/dot.py | 51 +-
cmake/FirstClassLangCuda.cmake | 50 +-
cpp-package/README.md | 11 +-
cpp-package/example/charRNN.cpp | 2 -
cpp-package/example/feature_extract/README.md | 8 +
cpp-package/include/mxnet-cpp/initializer.h | 6 +
cpp-package/include/mxnet-cpp/optimizer.h | 20 +-
cpp-package/include/mxnet-cpp/optimizer.hpp | 65 +-
dmlc-core | 2 +-
docs/_static/jquery-1.11.1.js | 10329 -------------------
docs/_static/js/page.js | 32 +-
docs/_static/mxnet-theme/index.html | 24 +-
docs/_static/mxnet-theme/layout.html | 5 +-
docs/_static/mxnet.css | 8 +-
docs/api/python/gluon/model_zoo.md | 35 +
docs/api/python/index.md | 9 +
docs/api/python/ndarray/sparse.md | 3 +-
docs/api/python/text/text.md | 455 +
docs/api/scala/index.md | 20 +-
docs/api/scala/io.md | 6 +-
docs/api/scala/kvstore.md | 5 +-
docs/api/scala/model.md | 4 +-
docs/api/scala/module.md | 10 +-
docs/api/scala/ndarray.md | 26 +-
docs/api/scala/symbol.md | 16 +-
docs/build_version_doc/build_all_version.sh | 2 +-
docs/community/contribute.md | 7 +-
docs/community/index.md | 8 +-
docs/faq/gradient_compression.md | 107 +
docs/faq/index.md | 12 +-
docs/faq/model_parallel_lstm.md | 12 +-
docs/faq/multi_devices.md | 13 +
docs/faq/torch.md | 62 -
docs/how_to/security.md | 24 +
docs/install/build_from_source.md | 37 +
docs/install/index.md | 376 +-
docs/install/windows_setup.md | 28 +-
docs/mxdoc.py | 4 +-
docs/tutorials/basic/image_io.md | 19 +-
docs/tutorials/basic/module.md | 91 +-
docs/tutorials/basic/ndarray_indexing.md | 377 +
docs/tutorials/c++/basics.md | 17 +-
.../tutorials/general_ml/recommendation_systems.md | 6 -
docs/tutorials/gluon/mnist.md | 2 +-
docs/tutorials/index.md | 19 +-
docs/tutorials/nlp/cnn.md | 162 +-
docs/tutorials/nlp/nce_loss.md | 38 -
docs/tutorials/nlp/rnn.md | 14 -
docs/tutorials/python/linear-regression.md | 5 +-
docs/tutorials/python/mnist.md | 9 +-
docs/tutorials/scala/char_lstm.md | 44 +-
docs/tutorials/scala/mnist.md | 4 +-
docs/tutorials/scala/mxnet_scala_on_intellij.md | 2 +-
docs/tutorials/sparse/csr.md | 181 +
docs/tutorials/sparse/row_sparse.md | 181 +-
docs/tutorials/sparse/train.md | 68 +-
.../tutorials/speech_recognition/baidu_warp_ctc.md | 97 -
docs/tutorials/speech_recognition/ctc.md | 15 +
docs/tutorials/speech_recognition/speech_lstm.md | 156 -
docs/tutorials/unsupervised_learning/gan.md | 162 +-
example/README.md | 98 +-
example/adversary/adversary_generation.ipynb | 7 +-
example/autoencoder/README.md | 16 +
example/autoencoder/autoencoder.py | 101 +-
example/autoencoder/data.py | 4 +
example/autoencoder/mnist_sae.py | 40 +-
example/autoencoder/model.py | 15 +-
example/autoencoder/solver.py | 47 +-
example/bayesian-methods/algos.py | 3 +-
example/bayesian-methods/bdk.ipynb | 20 +-
example/bayesian-methods/data_loader.py | 5 +-
example/bayesian-methods/sgld.ipynb | 8 +-
example/bi-lstm-sort/README.md | 48 +-
example/bi-lstm-sort/infer_sort.py | 25 +-
example/bi-lstm-sort/lstm.py | 1 -
example/bi-lstm-sort/lstm_sort.py | 75 +-
example/bi-lstm-sort/rnn_model.py | 1 -
example/bi-lstm-sort/sort_io.py | 1 -
example/caffe/caffe_net.py | 4 +-
example/caffe/data.py | 9 +-
example/caffe/train_model.py | 24 +-
example/capsnet/README.md | 66 +
example/capsnet/capsulelayers.py | 106 +
example/capsnet/capsulenet.py | 348 +
example/capsnet/result.PNG | Bin 0 -> 31313 bytes
example/captcha/mxnet_captcha.R | 8 +-
.../cnn_chinese_text_classification/text_cnn.py | 10 +-
example/cnn_text_classification/old/text_cnn.py | 279 -
example/cnn_text_classification/text_cnn.py | 4 +-
example/ctc/README.md | 227 +-
example/ctc/captcha_generator.py | 214 +
example/ctc/ctc_metrics.py | 114 +
example/ctc/hyperparams.py | 78 +
example/ctc/lstm.py | 120 +-
example/ctc/lstm_ocr.py | 254 -
example/ctc/lstm_ocr_infer.py | 93 +
example/ctc/lstm_ocr_train.py | 125 +
example/ctc/multiproc_data.py | 144 +
example/ctc/ocr_iter.py | 112 +
example/ctc/ocr_predict.py | 107 +-
example/{warpctc => ctc}/sample.jpg | Bin
example/dec/dec.py | 174 -
example/deep-embedded-clustering/README.md | 9 +
example/deep-embedded-clustering/autoencoder.py | 206 +
example/deep-embedded-clustering/data.py | 40 +
example/deep-embedded-clustering/dec.py | 174 +
example/deep-embedded-clustering/model.py | 78 +
example/deep-embedded-clustering/solver.py | 151 +
example/fcn-xs/README.md | 50 +-
example/fcn-xs/image_segmentaion.py | 109 +-
example/gan/dcgan.py | 299 -
example/gluon/data.py | 53 +-
example/gluon/image_classification.py | 57 +-
example/gluon/style_transfer/main.py | 4 +-
example/gluon/style_transfer/net.py | 4 +-
example/gluon/tree_lstm/main.py | 16 +-
example/gluon/word_language_model/README.md | 22 +-
example/gluon/word_language_model/get_ptb_data.sh | 11 +
.../word_language_model/get_wikitext2_data.sh | 36 +
example/gluon/word_language_model/train.py | 12 +-
example/image-classification/benchmark_score.py | 5 +-
example/image-classification/common/fit.py | 44 +-
example/image-classification/common/modelzoo.py | 2 +-
.../predict-cpp/image-classification-predict.cc | 19 -
example/image-classification/symbols/mobilenet.py | 156 +-
example/model-parallel-lstm/README.md | 4 -
example/model-parallel-lstm/get_ptb_data.sh | 32 -
example/model-parallel-lstm/lstm.py | 520 -
example/model-parallel-lstm/lstm_ptb.py | 128 -
example/model-parallel/lstm/README.md | 13 +
example/model-parallel/lstm/get_ptb_data.sh | 43 +
example/model-parallel/lstm/lstm.py | 532 +
example/model-parallel/lstm/lstm_ptb.py | 128 +
.../model-parallel/matrix_factorization/README.md | 22 +
.../matrix_factorization/get_data.py | 56 +
.../model-parallel/matrix_factorization/model.py | 57 +
.../model-parallel/matrix_factorization/train.py | 109 +
example/module/README.md | 9 +
example/module/lstm_bucketing.py | 103 -
example/module/mnist_mlp.py | 1 -
example/module/python_loss.py | 2 +-
example/module/sequential_module.py | 1 -
example/module/train_cifar10.ipynb | 593 --
example/module/train_cifar10.py | 215 -
example/multi-task/README.md | 4 +-
example/multi-task/example_multi_task.py | 24 +-
example/mxnet_adversarial_vae/README.md | 52 +-
example/mxnet_adversarial_vae/convert_data.py | 63 +-
example/mxnet_adversarial_vae/vaegan_mxnet.py | 525 +-
example/nce-loss/README.md | 63 +-
example/nce-loss/lstm_net.py | 104 +
example/nce-loss/lstm_word.py | 227 +-
example/nce-loss/nce.py | 67 +-
example/nce-loss/random_data.py | 127 +
example/nce-loss/text8_data.py | 362 +
example/nce-loss/toy_nce.py | 125 +-
example/nce-loss/toy_softmax.py | 105 +-
example/nce-loss/wordvec.py | 159 +-
example/nce-loss/wordvec_net.py | 81 +
example/nce-loss/wordvec_subwords.py | 290 +-
example/neural-style/end_to_end/README.md | 15 +-
example/neural-style/end_to_end/basic.py | 4 -
example/neural-style/end_to_end/boost_inference.py | 13 +-
example/neural-style/end_to_end/boost_train.py | 13 +-
example/neural-style/end_to_end/gen_v3.py | 10 -
example/neural-style/end_to_end/gen_v4.py | 10 -
example/numpy-ops/custom_softmax.py | 20 +-
example/numpy-ops/ndarray_softmax.py | 9 +-
example/numpy-ops/numpy_softmax.py | 20 +-
example/numpy-ops/weighted_logistic_regression.py | 40 +-
example/profiler/README.md | 23 +
example/profiler/profiler_executor.py | 2 +-
example/profiler/profiler_imageiter.py | 6 +-
example/profiler/profiler_matmul.py | 7 +-
example/profiler/profiler_ndarray.py | 12 +-
example/python-howto/README.md | 8 +-
example/python-howto/monitor_weights.py | 27 +-
example/rcnn/README.md | 22 +-
example/rcnn/rcnn/core/tester.py | 18 +-
example/rcnn/rcnn/cython/setup.py | 2 +-
example/rcnn/rcnn/dataset/__init__.py | 6 +-
example/rcnn/rcnn/dataset/coco.py | 18 +-
example/rcnn/rcnn/dataset/ds_utils.py | 2 +-
example/rcnn/rcnn/dataset/imdb.py | 7 +-
example/rcnn/rcnn/dataset/pascal_voc.py | 19 +-
example/rcnn/rcnn/dataset/pascal_voc_eval.py | 10 +-
example/rcnn/rcnn/io/rcnn.py | 6 +-
example/rcnn/rcnn/processing/bbox_regression.py | 2 +-
example/rcnn/rcnn/processing/generate_anchor.py | 4 +-
example/rcnn/rcnn/pycocotools/cocoeval.py | 4 +-
example/rcnn/rcnn/pycocotools/mask.py | 2 +-
example/rcnn/rcnn/symbol/__init__.py | 4 +-
example/rcnn/rcnn/symbol/proposal_target.py | 2 +-
example/rcnn/rcnn/symbol/symbol_resnet.py | 4 +-
example/rcnn/rcnn/symbol/symbol_vgg.py | 5 +-
example/rcnn/rcnn/tools/reeval.py | 7 +-
example/rcnn/script/additional_deps.sh | 14 +-
example/recommenders/README.md | 7 +
example/recommenders/movielens_data.py | 4 +-
example/recommenders/negativesample.py | 2 +-
example/reinforcement-learning/dqn/README.md | Bin 1032 -> 1752 bytes
example/reinforcement-learning/dqn/atari_game.py | 12 +-
example/reinforcement-learning/dqn/base.py | 2 +-
example/reinforcement-learning/dqn/dqn_demo.py | 4 +-
.../parallel_actor_critic/README.md | 8 +
.../parallel_actor_critic/model.py | 2 +-
.../parallel_actor_critic/train.py | 2 +-
example/rnn-time-major/bucket_io.py | 5 +-
example/rnn-time-major/get_ptb_data.sh | 11 +
example/rnn-time-major/readme.md | 24 +
example/rnn-time-major/rnn_cell_demo.py | 29 +-
example/rnn/README.md | 21 +-
example/rnn/bucket_R/aclImdb_lstm_classification.R | 33 +-
example/rnn/bucket_R/data_preprocessing.R | 170 -
.../rnn/bucket_R/data_preprocessing_seq_to_one.R | 8 +-
example/rnn/bucket_R/gru.cell.R | 54 -
example/rnn/bucket_R/lstm.cell.R | 41 -
example/rnn/bucket_R/mx.io.bucket.iter.R | 92 -
example/rnn/bucket_R/rnn.R | 208 -
example/rnn/bucket_R/rnn.infer.R | 79 -
example/rnn/bucket_R/rnn.train.R | 206 -
example/rnn/bucketing/README.md | 36 +
example/rnn/bucketing/cudnn_lstm_bucketing.py | 234 +
example/rnn/bucketing/get_ptb_data.sh | 43 +
example/rnn/bucketing/lstm_bucketing.py | 126 +
example/rnn/cudnn_lstm_bucketing.py | 234 -
example/rnn/get_ptb_data.sh | 32 -
example/rnn/lstm_bucketing.py | 127 -
example/rnn/old/get_ptb_data.sh | 11 +
example/rnn/word_lm/README.md | 49 +
example/rnn/word_lm/data.py | 114 +
example/rnn/word_lm/get_ptb_data.sh | 43 +
example/rnn/word_lm/model.py | 67 +
example/rnn/word_lm/module.py | 134 +
example/rnn/word_lm/train.py | 136 +
example/sparse/factorization_machine/README.md | 16 +
example/sparse/factorization_machine/metric.py | 88 +
example/sparse/factorization_machine/model.py | 54 +
example/sparse/factorization_machine/train.py | 142 +
example/sparse/get_data.py | 91 -
example/sparse/linear_classification.py | 133 -
example/sparse/linear_classification/README.md | 17 +
example/sparse/linear_classification/data.py | 33 +
.../{ => linear_classification}/linear_model.py | 0
example/sparse/linear_classification/train.py | 133 +
.../linear_classification/weighted_softmax_ce.py | 85 +
example/sparse/matrix_factorization.py | 114 -
example/sparse/matrix_factorization/README.md | 8 +
example/sparse/matrix_factorization/data.py | 62 +
.../model.py} | 0
example/sparse/matrix_factorization/train.py | 114 +
example/sparse/readme.md | 21 -
example/sparse/weighted_softmax_ce.py | 85 -
example/sparse/wide_deep/README.md | 7 +
example/sparse/wide_deep/data.py | 139 +
example/sparse/wide_deep/model.py | 58 +
example/sparse/wide_deep/train.py | 126 +
example/speech-demo/README.md | 160 -
example/speech-demo/config_util.py | 81 -
example/speech-demo/decode_mxnet.py | 177 -
example/speech-demo/decode_mxnet.sh | 112 -
example/speech-demo/default.cfg | 50 -
example/speech-demo/default_timit.cfg | 52 -
example/speech-demo/io_func/__init__.py | 0
example/speech-demo/io_func/convert2kaldi.py | 130 -
example/speech-demo/io_func/feat_io.py | 412 -
.../speech-demo/io_func/feat_readers/__init__.py | 0
example/speech-demo/io_func/feat_readers/common.py | 75 -
.../io_func/feat_readers/reader_atrack.py | 66 -
.../io_func/feat_readers/reader_bvec.py | 47 -
.../speech-demo/io_func/feat_readers/reader_htk.py | 54 -
.../io_func/feat_readers/reader_kaldi.py | 154 -
example/speech-demo/io_func/feat_readers/stats.py | 162 -
.../io_func/feat_readers/writer_kaldi.py | 74 -
example/speech-demo/io_func/info.py | 23 -
example/speech-demo/io_func/kaldi_parser.py | 219 -
example/speech-demo/io_func/model_io.py | 275 -
example/speech-demo/io_func/regr_feat_io.py | 92 -
example/speech-demo/io_func/utils.py | 170 -
example/speech-demo/io_util.py | 671 --
example/speech-demo/lstm_proj.py | 153 -
example/speech-demo/make_stats.py | 103 -
example/speech-demo/python_wrap/Makefile | 13 -
example/speech-demo/python_wrap/ctypes.cc | 244 -
.../python_wrap/example_usage/README.txt | 14 -
.../speech-demo/python_wrap/example_usage/data.ark | Bin 41 -> 0 bytes
.../speech-demo/python_wrap/example_usage/data.scp | 1 -
.../speech-demo/python_wrap/example_usage/data.txt | 3 -
.../python_wrap/example_usage/example.py | 111 -
example/speech-demo/run_ami.sh | 141 -
example/speech-demo/run_timit.sh | 141 -
example/speech-demo/speechSGD.py | 127 -
example/speech-demo/tests/test_nothing.py | 19 -
example/speech-demo/tests/test_system.py | 109 -
example/speech-demo/train_lstm_proj.py | 327 -
example/ssd/demo.py | 8 +-
example/ssd/deploy.py | 4 +-
example/ssd/evaluate.py | 8 +-
example/ssd/symbol/symbol_factory.py | 2 +-
example/ssd/tools/prepare_dataset.py | 4 +-
example/ssd/train.py | 10 +-
example/stochastic-depth/README.md | 29 +
example/stochastic-depth/sd_cifar10.py | 1 -
example/torch/torch_function.py | 32 -
example/torch/torch_module.py | 62 -
example/vae/README.md | 21 +
example/vae/VAE.py | 147 +-
example/vae/VAE_example.ipynb | 137 +-
example/warpctc/README.md | 108 -
example/warpctc/infer_ocr.py | 118 -
example/warpctc/lstm.py | 135 -
example/warpctc/lstm_model.py | 71 -
example/warpctc/lstm_ocr.py | 229 -
example/warpctc/ocr_predict.py | 106 -
example/warpctc/toy_ctc.py | 181 -
include/mxnet/base.h | 4 +-
include/mxnet/c_api.h | 43 +-
include/mxnet/imperative.h | 37 +-
include/mxnet/kvstore.h | 15 +
include/mxnet/ndarray.h | 7 +-
include/mxnet/op_attr_types.h | 2 -
include/mxnet/resource.h | 18 +-
include/mxnet/rtc.h | 4 +-
include/mxnet/storage.h | 16 +-
make/config.mk | 29 +-
make/osx.mk | 3 +
mshadow | 2 +-
nnvm | 2 +-
perl-package/AI-MXNet/README | 2 -
perl-package/AI-MXNet/examples/get_ptb_data.sh | 11 +
perl-package/AI-MXNet/lib/AI/MXNet.pm | 2 -
.../AI-MXNet/lib/AI/MXNet/Visualization.pm | 30 +-
perl-package/AI-MXNet/t/test_gluon_rnn.t | 18 +-
perl-package/AI-MXNet/t/test_loss.t | 2 +-
perl-package/AI-MXNet/t/test_random.t | 2 +-
perl-package/AI-MXNetCAPI/README | 2 -
perl-package/AI-MXNetCAPI/lib/AI/MXNetCAPI.pm | 2 -
perl-package/AI-NNVMCAPI/README | 2 -
perl-package/AI-NNVMCAPI/lib/AI/NNVMCAPI.pm | 2 -
perl-package/README.md | 1 -
perl-package/test.sh | 2 +-
plugin/opencv/cv_api.cc | 6 +-
prepare_mkl.sh | 8 +-
ps-lite | 2 +-
python/README.md | 21 +-
python/mxnet/_ctypes/ndarray.py | 7 +-
python/mxnet/autograd.py | 14 +-
python/mxnet/base.py | 2 +-
python/mxnet/contrib/__init__.py | 2 +
python/mxnet/contrib/text/__init__.py | 24 +
python/mxnet/contrib/text/_constants.py | 344 +
python/mxnet/contrib/text/embedding.py | 619 ++
python/mxnet/contrib/text/glossary.py | 118 +
python/mxnet/contrib/text/indexer.py | 218 +
python/mxnet/contrib/text/utils.py | 77 +
python/mxnet/gluon/block.py | 112 +-
python/mxnet/gluon/contrib/__init__.py | 2 +
python/mxnet/gluon/contrib/nn/__init__.py | 26 +
python/mxnet/gluon/contrib/nn/basic_layers.py | 112 +
python/mxnet/gluon/data/dataloader.py | 13 +-
python/mxnet/gluon/data/dataset.py | 4 +-
python/mxnet/gluon/data/vision/datasets.py | 16 +-
python/mxnet/gluon/loss.py | 28 +-
python/mxnet/gluon/model_zoo/custom_layers.py | 82 -
python/mxnet/gluon/model_zoo/model_store.py | 18 +-
python/mxnet/gluon/model_zoo/vision/__init__.py | 4 +-
python/mxnet/gluon/model_zoo/vision/alexnet.py | 5 +-
python/mxnet/gluon/model_zoo/vision/densenet.py | 9 +-
python/mxnet/gluon/model_zoo/vision/inception.py | 21 +-
python/mxnet/gluon/model_zoo/vision/mobilenet.py | 7 +-
python/mxnet/gluon/model_zoo/vision/resnet.py | 19 +-
python/mxnet/gluon/model_zoo/vision/squeezenet.py | 9 +-
python/mxnet/gluon/model_zoo/vision/vgg.py | 5 +-
python/mxnet/gluon/nn/basic_layers.py | 32 +-
python/mxnet/gluon/nn/conv_layers.py | 13 +-
python/mxnet/gluon/parameter.py | 83 +-
python/mxnet/gluon/rnn/rnn_cell.py | 24 +-
python/mxnet/gluon/trainer.py | 17 +-
python/mxnet/gluon/utils.py | 3 +-
python/mxnet/image/detection.py | 20 +-
python/mxnet/io.py | 27 +-
python/mxnet/kvstore.py | 89 +-
python/mxnet/libinfo.py | 6 +-
python/mxnet/model.py | 32 +-
python/mxnet/module/bucketing_module.py | 20 +-
python/mxnet/module/executor_group.py | 37 +-
python/mxnet/module/module.py | 14 +-
python/mxnet/ndarray/ndarray.py | 34 +-
python/mxnet/ndarray/sparse.py | 3 +-
python/mxnet/operator.py | 165 +-
python/mxnet/optimizer.py | 187 +-
python/mxnet/recordio.py | 2 +-
python/mxnet/registry.py | 17 +
python/mxnet/symbol/register.py | 24 +-
python/mxnet/symbol/symbol.py | 32 +-
python/mxnet/test_utils.py | 361 +-
python/setup.py | 4 +-
scala-package/README.md | 25 +-
scala-package/assembly/linux-x86_64-cpu/pom.xml | 12 +-
.../src/main/assembly/assembly.xml | 2 +-
scala-package/assembly/linux-x86_64-gpu/pom.xml | 12 +-
.../src/main/assembly/assembly.xml | 2 +-
scala-package/assembly/osx-x86_64-cpu/pom.xml | 12 +-
.../osx-x86_64-cpu/src/main/assembly/assembly.xml | 2 +-
scala-package/assembly/pom.xml | 4 +-
scala-package/core/pom.xml | 12 +-
.../src/main/scala/ml/dmlc/mxnet/AttrScope.scala | 55 -
.../core/src/main/scala/ml/dmlc/mxnet/Base.scala | 155 -
.../src/main/scala/ml/dmlc/mxnet/Callback.scala | 59 -
.../src/main/scala/ml/dmlc/mxnet/Context.scala | 83 -
.../core/src/main/scala/ml/dmlc/mxnet/DType.scala | 36 -
.../src/main/scala/ml/dmlc/mxnet/EvalMetric.scala | 325 -
.../src/main/scala/ml/dmlc/mxnet/Executor.scala | 309 -
.../main/scala/ml/dmlc/mxnet/ExecutorManager.scala | 541 -
.../src/main/scala/ml/dmlc/mxnet/FeedForward.scala | 685 --
.../core/src/main/scala/ml/dmlc/mxnet/IO.scala | 254 -
.../src/main/scala/ml/dmlc/mxnet/Initializer.scala | 171 -
.../src/main/scala/ml/dmlc/mxnet/KVStore.scala | 309 -
.../main/scala/ml/dmlc/mxnet/KVStoreServer.scala | 114 -
.../src/main/scala/ml/dmlc/mxnet/LRScheduler.scala | 70 -
.../src/main/scala/ml/dmlc/mxnet/LibInfo.scala | 321 -
.../core/src/main/scala/ml/dmlc/mxnet/Model.scala | 375 -
.../src/main/scala/ml/dmlc/mxnet/Monitor.scala | 132 -
.../src/main/scala/ml/dmlc/mxnet/NDArray.scala | 1213 ---
.../src/main/scala/ml/dmlc/mxnet/NameManager.scala | 67 -
.../src/main/scala/ml/dmlc/mxnet/Operator.scala | 276 -
.../src/main/scala/ml/dmlc/mxnet/Optimizer.scala | 316 -
.../src/main/scala/ml/dmlc/mxnet/Profiler.scala | 58 -
.../core/src/main/scala/ml/dmlc/mxnet/Random.scala | 95 -
.../src/main/scala/ml/dmlc/mxnet/RecordIO.scala | 231 -
.../core/src/main/scala/ml/dmlc/mxnet/Rtc.scala | 95 -
.../src/main/scala/ml/dmlc/mxnet/Serializer.scala | 75 -
.../core/src/main/scala/ml/dmlc/mxnet/Shape.scala | 59 -
.../core/src/main/scala/ml/dmlc/mxnet/Symbol.scala | 1314 ---
.../main/scala/ml/dmlc/mxnet/Visualization.scala | 323 -
.../main/scala/ml/dmlc/mxnet/contrib/NDArray.scala | 25 -
.../main/scala/ml/dmlc/mxnet/contrib/Symbol.scala | 25 -
.../main/scala/ml/dmlc/mxnet/io/MXDataIter.scala | 181 -
.../main/scala/ml/dmlc/mxnet/io/NDArrayIter.scala | 211 -
.../scala/ml/dmlc/mxnet/io/PrefetchingIter.scala | 181 -
.../main/scala/ml/dmlc/mxnet/io/ResizeIter.scala | 144 -
.../scala/ml/dmlc/mxnet/module/BaseModule.scala | 673 --
.../ml/dmlc/mxnet/module/BucketingModule.scala | 404 -
.../mxnet/module/DataParallelExecutorGroup.scala | 735 --
.../main/scala/ml/dmlc/mxnet/module/Module.scala | 645 --
.../ml/dmlc/mxnet/module/SequentialModule.scala | 409 -
.../scala/ml/dmlc/mxnet/optimizer/AdaDelta.scala | 111 -
.../scala/ml/dmlc/mxnet/optimizer/AdaGrad.scala | 88 -
.../main/scala/ml/dmlc/mxnet/optimizer/Adam.scala | 154 -
.../scala/ml/dmlc/mxnet/optimizer/DCASGD.scala | 123 -
.../main/scala/ml/dmlc/mxnet/optimizer/NAG.scala | 123 -
.../scala/ml/dmlc/mxnet/optimizer/RMSProp.scala | 116 -
.../main/scala/ml/dmlc/mxnet/optimizer/SGD.scala | 119 -
.../main/scala/ml/dmlc/mxnet/optimizer/SGLD.scala | 94 -
.../ml/dmlc/mxnet/util/NativeLibraryLoader.scala | 150 -
.../scala/ml/dmlc/mxnet/util/SerializerUtils.scala | 83 -
.../main/scala/org/apache/mxnet/AttrScope.scala | 55 +
.../src/main/scala/org/apache/mxnet/Base.scala | 155 +
.../src/main/scala/org/apache/mxnet/Callback.scala | 59 +
.../src/main/scala/org/apache/mxnet/Context.scala | 83 +
.../src/main/scala/org/apache/mxnet/DType.scala | 36 +
.../main/scala/org/apache/mxnet/EvalMetric.scala | 324 +
.../src/main/scala/org/apache/mxnet/Executor.scala | 305 +
.../scala/org/apache/mxnet/ExecutorManager.scala | 541 +
.../main/scala/org/apache/mxnet/FeedForward.scala | 685 ++
.../core/src/main/scala/org/apache/mxnet/IO.scala | 254 +
.../main/scala/org/apache/mxnet/Initializer.scala | 171 +
.../src/main/scala/org/apache/mxnet/KVStore.scala | 303 +
.../scala/org/apache/mxnet/KVStoreServer.scala | 114 +
.../main/scala/org/apache/mxnet/LRScheduler.scala | 70 +
.../src/main/scala/org/apache/mxnet/LibInfo.scala | 321 +
.../src/main/scala/org/apache/mxnet/Model.scala | 375 +
.../src/main/scala/org/apache/mxnet/Monitor.scala | 132 +
.../src/main/scala/org/apache/mxnet/NDArray.scala | 1207 +++
.../main/scala/org/apache/mxnet/NameManager.scala | 67 +
.../src/main/scala/org/apache/mxnet/Operator.scala | 276 +
.../main/scala/org/apache/mxnet/Optimizer.scala | 316 +
.../src/main/scala/org/apache/mxnet/Profiler.scala | 58 +
.../src/main/scala/org/apache/mxnet/Random.scala | 95 +
.../src/main/scala/org/apache/mxnet/RecordIO.scala | 231 +
.../core/src/main/scala/org/apache/mxnet/Rtc.scala | 95 +
.../main/scala/org/apache/mxnet/Serializer.scala | 75 +
.../src/main/scala/org/apache/mxnet/Shape.scala | 59 +
.../src/main/scala/org/apache/mxnet/Symbol.scala | 1308 +++
.../scala/org/apache/mxnet/Visualization.scala | 323 +
.../scala/org/apache/mxnet/contrib/NDArray.scala | 25 +
.../scala/org/apache/mxnet/contrib/Symbol.scala | 25 +
.../scala/org/apache/mxnet/io/MXDataIter.scala | 179 +
.../scala/org/apache/mxnet/io/NDArrayIter.scala | 215 +
.../org/apache/mxnet/io/PrefetchingIter.scala | 181 +
.../scala/org/apache/mxnet/io/ResizeIter.scala | 144 +
.../scala/org/apache/mxnet/module/BaseModule.scala | 684 ++
.../org/apache/mxnet/module/BucketingModule.scala | 404 +
.../mxnet/module/DataParallelExecutorGroup.scala | 747 ++
.../scala/org/apache/mxnet/module/Module.scala | 645 ++
.../org/apache/mxnet/module/SequentialModule.scala | 409 +
.../org/apache/mxnet/optimizer/AdaDelta.scala | 111 +
.../scala/org/apache/mxnet/optimizer/AdaGrad.scala | 88 +
.../scala/org/apache/mxnet/optimizer/Adam.scala | 154 +
.../scala/org/apache/mxnet/optimizer/DCASGD.scala | 123 +
.../scala/org/apache/mxnet/optimizer/NAG.scala | 123 +
.../scala/org/apache/mxnet/optimizer/RMSProp.scala | 116 +
.../scala/org/apache/mxnet/optimizer/SGD.scala | 119 +
.../scala/org/apache/mxnet/optimizer/SGLD.scala | 94 +
.../apache/mxnet/util/NativeLibraryLoader.scala | 150 +
.../org/apache/mxnet/util/SerializerUtils.scala | 83 +
.../org/apache/mxnet/util/WarnIfNotDisposed.scala | 88 +
.../test/scala/ml/dmlc/mxnet/AttrScopeSuite.scala | 36 -
.../src/test/scala/ml/dmlc/mxnet/CheckUtils.scala | 33 -
.../test/scala/ml/dmlc/mxnet/ExecutorSuite.scala | 80 -
.../src/test/scala/ml/dmlc/mxnet/IOSuite.scala | 289 -
.../test/scala/ml/dmlc/mxnet/KVStoreSuite.scala | 101 -
.../scala/ml/dmlc/mxnet/ModelParallelSuite.scala | 78 -
.../src/test/scala/ml/dmlc/mxnet/ModuleSuite.scala | 368 -
.../test/scala/ml/dmlc/mxnet/NDArraySuite.scala | 578 --
.../test/scala/ml/dmlc/mxnet/OperatorSuite.scala | 1069 --
.../src/test/scala/ml/dmlc/mxnet/RandomSuite.scala | 53 -
.../test/scala/ml/dmlc/mxnet/RecordIOSuite.scala | 86 -
.../test/scala/ml/dmlc/mxnet/SerializerSuite.scala | 48 -
.../src/test/scala/ml/dmlc/mxnet/ShapeSuite.scala | 32 -
.../src/test/scala/ml/dmlc/mxnet/SymbolSuite.scala | 74 -
.../src/test/scala/ml/dmlc/mxnet/TestUtil.scala | 47 -
.../test/scala/ml/dmlc/mxnet/gpu/RtcSuite.scala | 64 -
.../test/scala/ml/dmlc/mxnet/train/ConvSuite.scala | 115 -
.../ml/dmlc/mxnet/util/SerializerUtilsSuite.scala | 33 -
.../scala/org/apache/mxnet/AttrScopeSuite.scala | 36 +
.../test/scala/org/apache/mxnet/CheckUtils.scala | 33 +
.../scala/org/apache/mxnet/ExecutorSuite.scala | 80 +
.../src/test/scala/org/apache/mxnet/IOSuite.scala | 289 +
.../test/scala/org/apache/mxnet/KVStoreSuite.scala | 101 +
.../org/apache/mxnet/ModelParallelSuite.scala | 78 +
.../test/scala/org/apache/mxnet/ModuleSuite.scala | 368 +
.../test/scala/org/apache/mxnet/NDArraySuite.scala | 578 ++
.../scala/org/apache/mxnet/OperatorSuite.scala | 1069 ++
.../test/scala/org/apache/mxnet/RandomSuite.scala | 53 +
.../scala/org/apache/mxnet/RecordIOSuite.scala | 86 +
.../scala/org/apache/mxnet/SerializerSuite.scala | 48 +
.../test/scala/org/apache/mxnet/ShapeSuite.scala | 32 +
.../test/scala/org/apache/mxnet/SymbolSuite.scala | 74 +
.../src/test/scala/org/apache/mxnet/TestUtil.scala | 47 +
.../test/scala/org/apache/mxnet/gpu/RtcSuite.scala | 64 +
.../scala/org/apache/mxnet/train/ConvSuite.scala | 115 +
.../apache/mxnet/util/SerializerUtilsSuite.scala | 33 +
.../apache/mxnet/util/WarnIfNotDiposedSuite.scala | 63 +
scala-package/dev/change-artifact-id.sh | 2 +-
scala-package/dev/change-scala-version.sh | 5 +-
scala-package/examples/pom.xml | 14 +-
.../examples/scripts/customop/run_customop.sh | 2 +-
.../scripts/customop/run_customopwithrtc.sh | 2 +-
scala-package/examples/scripts/module/mnist_mlp.sh | 2 +-
.../scripts/module/run_sequential_module.sh | 2 +-
.../neuralstyle_end2end/run_test_end2end.sh | 2 +-
.../neuralstyle_end2end/run_train_end2end.sh | 2 +-
.../scripts/profiler/run_profiler_matmul.sh | 2 +-
.../scripts/profiler/run_profiler_ndarray.sh | 2 +-
.../examples/scripts/rnn/run_lstm_bucketing.sh | 2 +-
.../examples/scripts/rnn/run_test_charrnn.sh | 2 +-
.../examples/scripts/rnn/run_train_charrnn.sh | 2 +-
.../examples/scripts/run_cnntextclassification.sh | 2 +-
scala-package/examples/scripts/run_gan_mnist.sh | 2 +-
scala-package/examples/scripts/run_multitask.sh | 2 +-
scala-package/examples/scripts/run_neuralstyle.sh | 2 +-
.../examples/scripts/run_visualization.sh | 2 +-
.../CNNTextClassification.scala | 288 -
.../cnntextclassification/DataHelper.scala | 173 -
.../ml/dmlc/mxnetexamples/customop/Data.scala | 50 -
.../mxnetexamples/customop/ExampleCustomOp.scala | 225 -
.../customop/ExampleCustomOpWithRtc.scala | 244 -
.../scala/ml/dmlc/mxnetexamples/gan/GanMnist.scala | 215 -
.../scala/ml/dmlc/mxnetexamples/gan/Module.scala | 165 -
.../main/scala/ml/dmlc/mxnetexamples/gan/Viz.scala | 106 -
.../imclassification/ModelTrain.scala | 112 -
.../imclassification/TrainMnist.scala | 195 -
.../ml/dmlc/mxnetexamples/module/MnistMlp.scala | 171 -
.../mxnetexamples/module/SequentialModuleEx.scala | 191 -
.../ml/dmlc/mxnetexamples/multitask/Data.scala | 51 -
.../mxnetexamples/multitask/ExampleMultiTask.scala | 336 -
.../mxnetexamples/neuralstyle/ModelVgg19.scala | 133 -
.../mxnetexamples/neuralstyle/NeuralStyle.scala | 298 -
.../mxnetexamples/neuralstyle/end2end/Basic.scala | 123 -
.../neuralstyle/end2end/BoostInference.scala | 87 -
.../neuralstyle/end2end/BoostTrain.scala | 208 -
.../neuralstyle/end2end/DataProcessing.scala | 94 -
.../mxnetexamples/neuralstyle/end2end/GenV3.scala | 85 -
.../mxnetexamples/neuralstyle/end2end/GenV4.scala | 107 -
.../neuralstyle/end2end/ModelVgg19.scala | 111 -
.../mxnetexamples/neuralstyle/end2end/Module.scala | 148 -
.../mxnetexamples/profiler/ProfilerMatMul.scala | 105 -
.../mxnetexamples/profiler/ProfilerNDArray.scala | 250 -
.../scala/ml/dmlc/mxnetexamples/rnn/BucketIo.scala | 248 -
.../scala/ml/dmlc/mxnetexamples/rnn/Lstm.scala | 170 -
.../ml/dmlc/mxnetexamples/rnn/LstmBucketing.scala | 131 -
.../scala/ml/dmlc/mxnetexamples/rnn/RnnModel.scala | 73 -
.../ml/dmlc/mxnetexamples/rnn/TestCharRnn.scala | 105 -
.../ml/dmlc/mxnetexamples/rnn/TrainCharRnn.scala | 187 -
.../scala/ml/dmlc/mxnetexamples/rnn/Utils.scala | 154 -
.../dmlc/mxnetexamples/visualization/AlexNet.scala | 72 -
.../mxnetexamples/visualization/ExampleVis.scala | 90 -
.../mxnetexamples/visualization/GoogleNet.scala | 94 -
.../mxnetexamples/visualization/Inception_BN.scala | 127 -
.../mxnetexamples/visualization/Inception_V3.scala | 253 -
.../dmlc/mxnetexamples/visualization/LeNet.scala | 52 -
.../mxnetexamples/visualization/ResNet_Small.scala | 110 -
.../ml/dmlc/mxnetexamples/visualization/VGG.scala | 84 -
.../CNNTextClassification.scala | 288 +
.../cnntextclassification/DataHelper.scala | 173 +
.../org/apache/mxnetexamples/customop/Data.scala | 50 +
.../mxnetexamples/customop/ExampleCustomOp.scala | 225 +
.../customop/ExampleCustomOpWithRtc.scala | 244 +
.../org/apache/mxnetexamples/gan/GanMnist.scala | 215 +
.../org/apache/mxnetexamples/gan/Module.scala | 165 +
.../scala/org/apache/mxnetexamples/gan/Viz.scala | 106 +
.../imclassification/ModelTrain.scala | 112 +
.../imclassification/TrainMnist.scala | 195 +
.../org/apache/mxnetexamples/module/MnistMlp.scala | 171 +
.../mxnetexamples/module/SequentialModuleEx.scala | 191 +
.../org/apache/mxnetexamples/multitask/Data.scala | 51 +
.../mxnetexamples/multitask/ExampleMultiTask.scala | 336 +
.../mxnetexamples/neuralstyle/ModelVgg19.scala | 133 +
.../mxnetexamples/neuralstyle/NeuralStyle.scala | 298 +
.../mxnetexamples/neuralstyle/end2end/Basic.scala | 123 +
.../neuralstyle/end2end/BoostInference.scala | 87 +
.../neuralstyle/end2end/BoostTrain.scala | 208 +
.../neuralstyle/end2end/DataProcessing.scala | 94 +
.../mxnetexamples/neuralstyle/end2end/GenV3.scala | 85 +
.../mxnetexamples/neuralstyle/end2end/GenV4.scala | 107 +
.../neuralstyle/end2end/ModelVgg19.scala | 111 +
.../mxnetexamples/neuralstyle/end2end/Module.scala | 148 +
.../mxnetexamples/profiler/ProfilerMatMul.scala | 105 +
.../mxnetexamples/profiler/ProfilerNDArray.scala | 250 +
.../org/apache/mxnetexamples/rnn/BucketIo.scala | 248 +
.../scala/org/apache/mxnetexamples/rnn/Lstm.scala | 170 +
.../apache/mxnetexamples/rnn/LstmBucketing.scala | 131 +
.../org/apache/mxnetexamples/rnn/RnnModel.scala | 73 +
.../org/apache/mxnetexamples/rnn/TestCharRnn.scala | 105 +
.../apache/mxnetexamples/rnn/TrainCharRnn.scala | 187 +
.../scala/org/apache/mxnetexamples/rnn/Utils.scala | 154 +
.../mxnetexamples/visualization/AlexNet.scala | 72 +
.../mxnetexamples/visualization/ExampleVis.scala | 90 +
.../mxnetexamples/visualization/GoogleNet.scala | 94 +
.../mxnetexamples/visualization/Inception_BN.scala | 127 +
.../mxnetexamples/visualization/Inception_V3.scala | 253 +
.../apache/mxnetexamples/visualization/LeNet.scala | 52 +
.../mxnetexamples/visualization/ResNet_Small.scala | 110 +
.../apache/mxnetexamples/visualization/VGG.scala | 84 +
scala-package/init-native/linux-x86_64/pom.xml | 14 +-
scala-package/init-native/osx-x86_64/pom.xml | 14 +-
scala-package/init-native/pom.xml | 4 +-
.../main/native/ml_dmlc_mxnet_init_native_c_api.cc | 120 -
.../native/org_apache_mxnet_init_native_c_api.cc | 120 +
scala-package/init/pom.xml | 4 +-
.../src/main/scala/ml/dmlc/mxnet/init/Base.scala | 52 -
.../main/scala/ml/dmlc/mxnet/init/LibInfo.scala | 36 -
.../main/scala/org/apache/mxnet/init/Base.scala | 52 +
.../main/scala/org/apache/mxnet/init/LibInfo.scala | 36 +
scala-package/macros/pom.xml | 12 +-
.../main/scala/ml/dmlc/mxnet/NDArrayMacro.scala | 187 -
.../src/main/scala/ml/dmlc/mxnet/SymbolMacro.scala | 179 -
.../ml/dmlc/mxnet/utils/OperatorBuildUtils.scala | 32 -
.../main/scala/org/apache/mxnet/NDArrayMacro.scala | 187 +
.../main/scala/org/apache/mxnet/SymbolMacro.scala | 179 +
.../apache/mxnet/utils/OperatorBuildUtils.scala | 32 +
scala-package/native/linux-x86_64-cpu/pom.xml | 14 +-
scala-package/native/linux-x86_64-gpu/pom.xml | 14 +-
scala-package/native/osx-x86_64-cpu/pom.xml | 14 +-
scala-package/native/pom.xml | 4 +-
.../native/src/main/native/jni_helper_func.h | 10 +-
.../src/main/native/ml_dmlc_mxnet_native_c_api.cc | 2529 -----
.../main/native/org_apache_mxnet_native_c_api.cc | 2529 +++++
scala-package/pom.xml | 104 +-
scala-package/spark/bin/run-mnist-example.sh | 2 +-
scala-package/spark/pom.xml | 8 +-
.../main/scala/ml/dmlc/mxnet/spark/MXNDArray.scala | 40 -
.../src/main/scala/ml/dmlc/mxnet/spark/MXNet.scala | 263 -
.../scala/ml/dmlc/mxnet/spark/MXNetModel.scala | 86 -
.../scala/ml/dmlc/mxnet/spark/MXNetParams.scala | 69 -
.../ml/dmlc/mxnet/spark/ParameterServer.scala | 165 -
.../spark/example/ClassificationExample.scala | 190 -
.../ml/dmlc/mxnet/spark/io/LabeledPointIter.scala | 139 -
.../dmlc/mxnet/spark/io/LongLivingDataBatch.scala | 33 -
.../scala/ml/dmlc/mxnet/spark/io/PointIter.scala | 138 -
.../ml/dmlc/mxnet/spark/transformer/MXNet.scala | 172 -
.../ml/dmlc/mxnet/spark/utils/Img2Vector.scala | 85 -
.../scala/ml/dmlc/mxnet/spark/utils/Network.scala | 65 -
.../ml/dmlc/mxnet/spark/utils/RepIterator.scala | 49 -
.../scala/org/apache/mxnet/spark/MXNDArray.scala | 40 +
.../main/scala/org/apache/mxnet/spark/MXNet.scala | 263 +
.../scala/org/apache/mxnet/spark/MXNetModel.scala | 86 +
.../scala/org/apache/mxnet/spark/MXNetParams.scala | 69 +
.../org/apache/mxnet/spark/ParameterServer.scala | 165 +
.../spark/example/ClassificationExample.scala | 190 +
.../apache/mxnet/spark/io/LabeledPointIter.scala | 139 +
.../mxnet/spark/io/LongLivingDataBatch.scala | 33 +
.../org/apache/mxnet/spark/io/PointIter.scala | 138 +
.../org/apache/mxnet/spark/transformer/MXNet.scala | 172 +
.../org/apache/mxnet/spark/utils/Img2Vector.scala | 85 +
.../org/apache/mxnet/spark/utils/Network.scala | 65 +
.../org/apache/mxnet/spark/utils/RepIterator.scala | 49 +
setup-utils/install-mxnet-osx-python.sh | 103 +-
snapcraft.yaml | 2 +-
src/c_api/c_api.cc | 36 +-
src/c_api/c_api_function.cc | 72 +-
src/c_api/c_api_ndarray.cc | 32 +-
src/c_api/c_api_symbolic.cc | 5 +
src/common/cuda_utils.h | 5 +
src/common/random_generator.cu | 63 +
src/common/random_generator.h | 218 +
src/common/rtc.cc | 6 +-
src/common/utils.cc | 10 +
src/common/utils.cu | 10 +
src/common/utils.h | 38 +-
src/engine/profiler.cc | 50 +-
src/engine/profiler.h | 22 +-
src/engine/threaded_engine_perdevice.cc | 37 +-
src/executor/attach_op_resource_pass.cc | 2 +
src/executor/graph_executor.cc | 171 +-
src/executor/graph_executor.h | 4 +
src/imperative/cached_op.cc | 43 +-
src/imperative/imperative.cc | 16 +-
src/imperative/imperative_utils.h | 103 +-
src/initialize.cc | 26 +-
src/io/image_io.cc | 26 +-
src/kvstore/comm.h | 215 +-
src/kvstore/gradient_compression-inl.h | 155 +
src/kvstore/gradient_compression.cc | 193 +
src/kvstore/gradient_compression.cu | 40 +
src/kvstore/gradient_compression.h | 138 +
src/kvstore/kvstore.cc | 16 +-
src/kvstore/kvstore_dist.h | 388 +-
src/kvstore/kvstore_dist_server.h | 152 +-
src/kvstore/kvstore_local.h | 88 +-
src/kvstore/kvstore_nccl.h | 549 +
src/kvstore/kvstore_utils.cc | 45 +
src/kvstore/kvstore_utils.cu | 102 +
src/kvstore/kvstore_utils.h | 47 +
src/ndarray/ndarray.cc | 217 +-
src/ndarray/ndarray_function-inl.h | 2 +-
src/ndarray/ndarray_function.cc | 13 +
src/ndarray/ndarray_function.cu | 12 +
src/ndarray/ndarray_function.h | 33 +-
src/operator/activation-inl.h | 218 -
src/operator/activation.cc | 97 -
src/operator/activation.cu | 69 -
src/operator/batch_norm-inl.h | 493 -
src/operator/batch_norm.cc | 421 -
src/operator/batch_norm.cu | 666 --
src/operator/c_lapack_api.h | 14 +-
src/operator/concat-inl.h | 39 +-
src/operator/contrib/bounding_box-inl.h | 728 ++
src/operator/contrib/bounding_box.cc | 194 +
src/operator/contrib/bounding_box.cu | 50 +
src/operator/contrib/ctc_loss-inl.h | 2 +-
src/operator/contrib/deformable_convolution-inl.h | 10 +-
src/operator/contrib/krprod.cc | 136 +
src/operator/contrib/krprod.h | 40 +-
src/operator/contrib/nn/deformable_im2col.h | 19 -
src/operator/contrib/psroi_pooling-inl.h | 19 -
src/operator/convolution-inl.h | 684 --
src/operator/convolution.cc | 177 -
src/operator/convolution.cu | 88 -
src/operator/convolution_v1.cu | 2 +-
src/operator/cudnn_activation-inl.h | 228 -
src/operator/cudnn_algoreg-inl.h | 178 -
src/operator/cudnn_batch_norm-inl.h | 390 -
src/operator/cudnn_convolution-inl.h | 915 --
src/operator/cudnn_deconvolution-inl.h | 954 --
src/operator/cudnn_pooling-inl.h | 315 -
src/operator/cudnn_softmax_activation-inl.h | 184 -
src/operator/custom/custom-inl.h | 76 +-
src/operator/custom/custom.cc | 185 +-
src/operator/deconvolution-inl.h | 753 --
src/operator/deconvolution.cc | 65 -
src/operator/deconvolution.cu | 73 -
src/operator/depthwise_convolution-inl.h | 353 -
src/operator/depthwise_convolution_tf.cuh | 733 --
src/operator/dropout-inl.h | 288 -
src/operator/dropout.cc | 83 -
src/operator/fully_connected-inl.h | 300 -
src/operator/linalg_impl.h | 145 +-
src/operator/mkl/mkl_convolution-inl.h | 2 +-
src/operator/mkl/mkl_pooling-inl.h | 2 +-
src/operator/mshadow_op.h | 91 +-
src/operator/mxnet_op.h | 160 +-
src/operator/nn/activation-inl.h | 218 +
src/operator/nn/activation.cc | 97 +
src/operator/nn/activation.cu | 69 +
src/operator/nn/batch_norm-inl.h | 493 +
src/operator/nn/batch_norm.cc | 421 +
src/operator/nn/batch_norm.cu | 666 ++
src/operator/nn/convolution-inl.h | 684 ++
src/operator/nn/convolution.cc | 177 +
src/operator/nn/convolution.cu | 81 +
src/operator/nn/cudnn/cudnn_activation-inl.h | 228 +
src/operator/nn/cudnn/cudnn_algoreg-inl.h | 183 +
src/operator/{ => nn/cudnn}/cudnn_algoreg.cc | 0
src/operator/nn/cudnn/cudnn_batch_norm-inl.h | 390 +
src/operator/{ => nn/cudnn}/cudnn_batch_norm.cc | 0
src/operator/{ => nn/cudnn}/cudnn_batch_norm.cu | 0
src/operator/nn/cudnn/cudnn_convolution-inl.h | 886 ++
src/operator/nn/cudnn/cudnn_deconvolution-inl.h | 931 ++
src/operator/nn/cudnn/cudnn_pooling-inl.h | 315 +
.../nn/cudnn/cudnn_softmax_activation-inl.h | 184 +
src/operator/nn/deconvolution-inl.h | 788 ++
src/operator/nn/deconvolution.cc | 65 +
src/operator/nn/deconvolution.cu | 67 +
src/operator/nn/depthwise_convolution-inl.h | 353 +
src/operator/nn/depthwise_convolution_tf.cuh | 733 ++
src/operator/nn/dropout-inl.h | 402 +
src/operator/nn/dropout.cc | 84 +
src/operator/{ => nn}/dropout.cu | 0
src/operator/nn/fully_connected-inl.h | 300 +
src/operator/{ => nn}/fully_connected.cc | 0
src/operator/{ => nn}/fully_connected.cu | 0
src/operator/nn/pool.h | 19 -
src/operator/nn/pooling-inl.h | 335 +
src/operator/nn/pooling.cc | 138 +
src/operator/nn/pooling.cu | 69 +
src/operator/nn/softmax.cc | 2 +-
src/operator/nn/softmax.cu | 2 +-
src/operator/nn/softmax_activation-inl.h | 213 +
src/operator/nn/softmax_activation.cc | 72 +
src/operator/nn/softmax_activation.cu | 44 +
src/operator/nn/upsampling-inl.h | 335 +
src/operator/{ => nn}/upsampling.cc | 0
src/operator/{ => nn}/upsampling.cu | 0
src/operator/operator_common.h | 46 +-
src/operator/operator_tune-inl.h | 758 ++
src/operator/operator_tune.cc | 353 +
src/operator/operator_tune.h | 286 +
src/operator/optimizer_op-inl.h | 421 +-
src/operator/optimizer_op.cc | 218 +-
src/operator/optimizer_op.cu | 144 +
src/operator/pooling-inl.h | 335 -
src/operator/pooling.cc | 138 -
src/operator/pooling.cu | 69 -
src/operator/random/multisample_op.cc | 3 +-
src/operator/random/multisample_op.h | 23 +-
src/operator/random/sample_op.h | 108 +-
src/operator/random/sampler.h | 293 +-
src/operator/regression_output.cc | 6 +-
src/operator/regression_output.cu | 6 +-
src/operator/roi_pooling.cc | 2 +-
src/operator/sequence_last-inl.h | 178 +-
src/operator/sequence_mask-inl.h | 142 +-
src/operator/sequence_op_common.h | 4 +-
src/operator/sequence_reverse-inl.h | 40 +-
src/operator/softmax_activation-inl.h | 213 -
src/operator/softmax_activation.cc | 72 -
src/operator/softmax_activation.cu | 44 -
src/operator/special_functions-inl.h | 19 -
src/operator/tensor/broadcast_reduce-inl.h | 2 +
src/operator/tensor/broadcast_reduce_op.h | 6 +-
src/operator/tensor/broadcast_reduce_op_value.cc | 2 +-
src/operator/tensor/dot-inl.h | 213 +-
src/operator/tensor/dot.cc | 1 +
src/operator/tensor/elemwise_binary_broadcast_op.h | 46 +-
.../tensor/elemwise_binary_broadcast_op_basic.cc | 8 +-
.../tensor/elemwise_binary_broadcast_op_basic.cu | 8 +-
src/operator/tensor/elemwise_binary_op-inl.h | 10 +-
src/operator/tensor/elemwise_binary_op.h | 10 +-
src/operator/tensor/elemwise_binary_op_basic.cc | 12 +-
src/operator/tensor/elemwise_binary_op_basic.cu | 11 +-
src/operator/tensor/elemwise_binary_scalar_op.h | 2 +-
.../tensor/elemwise_binary_scalar_op_basic.cc | 53 +-
.../tensor/elemwise_binary_scalar_op_basic.cu | 20 +-
src/operator/tensor/elemwise_scatter_op.cc | 13 +-
src/operator/tensor/elemwise_scatter_op.cu | 13 +-
src/operator/tensor/elemwise_unary_op.h | 19 +-
src/operator/tensor/elemwise_unary_op_basic.cc | 15 +-
src/operator/tensor/indexing_op.cc | 127 +-
src/operator/tensor/indexing_op.cu | 37 +-
src/operator/tensor/indexing_op.h | 69 +-
src/operator/tensor/init_op.cu | 1 +
src/operator/tensor/init_op.h | 17 +-
src/operator/tensor/matrix_op-inl.h | 84 +-
src/operator/tensor/matrix_op.cc | 64 +-
src/operator/tensor/matrix_op.cu | 113 +-
src/operator/tensor/square_sum-inl.h | 87 +-
src/operator/tensor/square_sum.cc | 26 +
src/operator/tensor/square_sum.cu | 62 +
src/operator/upsampling-inl.h | 335 -
src/resource.cc | 97 +-
src/storage/cpu_shared_storage_manager.h | 76 +-
src/storage/gpu_device_storage.h | 7 +
src/storage/pinned_memory_storage.h | 7 +
src/storage/pooled_storage_manager.h | 15 +-
tests/CMakeLists.txt | 4 +-
tests/ci_build/Dockerfile.build_cuda | 26 +
tests/ci_build/Dockerfile.caffe_gpu | 13 +-
tests/ci_build/Dockerfile.cpu | 2 +-
tests/ci_build/Dockerfile.cpu_clang | 21 +
tests/ci_build/Dockerfile.cpu_mklml | 18 +
tests/ci_build/Dockerfile.gpu | 4 +-
tests/ci_build/Dockerfile.gpu_mklml | 18 +
tests/ci_build/Dockerfile.lint | 5 +-
tests/ci_build/Dockerfile.mklml_gpu | 15 -
tests/ci_build/ci_build.sh | 47 +-
tests/ci_build/install/ubuntu_install_core.sh | 5 +-
tests/ci_build/install/ubuntu_install_nvidia.sh | 31 +
tests/ci_build/pip_tests/Dockerfile.in.pip_cpu | 2 +-
.../ci_build/pip_tests/Dockerfile.pip_dependencies | 2 +-
tests/cpp/engine/threaded_engine_test.cc | 3 +-
tests/cpp/include/test_core_op.h | 16 +-
tests/cpp/include/test_legacy_op.h | 7 +
tests/cpp/include/test_op.h | 7 +-
tests/cpp/include/test_op_runner.h | 36 +-
tests/cpp/include/test_perf.h | 13 +
tests/cpp/include/test_tune.h | 338 +
tests/cpp/include/test_util.h | 28 +-
tests/cpp/misc/memory_test.cc | 34 +-
tests/cpp/operator/activation_perf.cc | 2 +-
tests/cpp/operator/batchnorm_test.cc | 216 +-
tests/cpp/operator/broadcast_perf.cc | 101 -
tests/cpp/operator/coreop_perf.cc | 2 +-
tests/cpp/operator/dropout_perf.cc | 104 +
tests/cpp/operator/fully_conn_perf.cc | 2 +-
tests/cpp/operator/tune/operator_tune_test.cc | 175 +
tests/cpp/test_main.cc | 16 +-
tests/cpp/unittest.mk | 17 +-
tests/nightly/dist_sync_kvstore.py | 120 +-
tests/nightly/test_kvstore.py | 200 +-
tests/python/common/get_data.py | 81 -
tests/python/cpu/test_mklml.py | 58 +
tests/python/gpu/test_kvstore_gpu.py | 75 +-
tests/python/gpu/test_nccl.py | 52 +
tests/python/gpu/test_operator_gpu.py | 77 +-
tests/python/train/test_autograd.py | 4 +-
tests/python/train/test_conv.py | 4 +-
tests/python/train/test_dtype.py | 4 +-
tests/python/train/test_mlp.py | 4 +-
tests/python/unittest/common.py | 2 -
tests/python/unittest/test_contrib_krprod.py | 65 +
tests/python/unittest/test_contrib_operator.py | 231 +
tests/python/unittest/test_contrib_text.py | 690 ++
tests/python/unittest/test_executor.py | 77 +-
tests/python/unittest/test_gluon.py | 111 +-
tests/python/unittest/test_gluon_contrib.py | 35 +
tests/python/unittest/test_gluon_model_zoo.py | 29 +-
tests/python/unittest/test_image.py | 284 +-
tests/python/unittest/test_init.py | 16 +
tests/python/unittest/test_io.py | 18 +-
tests/python/unittest/test_kvstore.py | 1 +
tests/python/unittest/test_loss.py | 21 +-
tests/python/unittest/test_module.py | 66 +-
tests/python/unittest/test_ndarray.py | 19 +-
tests/python/unittest/test_operator.py | 542 +-
tests/python/unittest/test_optimizer.py | 205 +-
tests/python/unittest/test_profiler.py | 11 +-
tests/python/unittest/test_random.py | 202 +-
tests/python/unittest/test_sparse_ndarray.py | 13 +-
tests/python/unittest/test_sparse_operator.py | 140 +-
tests/python/unittest/test_symbol.py | 6 +-
tools/bandwidth/measure.py | 6 +-
tools/caffe_converter/compare_layers.py | 10 +-
tools/caffe_converter/convert_caffe_modelzoo.py | 4 +-
tools/caffe_translator/README.md | 79 +
tools/caffe_translator/build.gradle | 152 +
tools/caffe_translator/build_from_source.md | 21 +
tools/caffe_translator/faq.md | 17 +
tools/caffe_translator/gradle.properties | 12 +
.../gradle/wrapper/gradle-wrapper.jar | Bin 0 -> 54731 bytes
.../gradle/wrapper/gradle-wrapper.properties | 5 +
tools/caffe_translator/gradlew | 189 +
tools/caffe_translator/gradlew.bat | 101 +
.../scripts/convert_caffe_model.py | 121 +
tools/caffe_translator/settings.gradle | 2 +
.../io/mxnet/caffetranslator/CaffePrototxt.g4 | 74 +
.../main/java/io/mxnet/caffetranslator/Config.java | 55 +
.../java/io/mxnet/caffetranslator/Converter.java | 429 +
.../mxnet/caffetranslator/CreateModelListener.java | 144 +
.../io/mxnet/caffetranslator/GenerationHelper.java | 195 +
.../io/mxnet/caffetranslator/GeneratorOutput.java | 35 +
.../java/io/mxnet/caffetranslator/Launcher.java | 178 +
.../main/java/io/mxnet/caffetranslator/Layer.java | 141 +
.../java/io/mxnet/caffetranslator/MLModel.java | 105 +
.../java/io/mxnet/caffetranslator/Optimizer.java | 48 +
.../io/mxnet/caffetranslator/ParserHelper.java | 36 +
.../main/java/io/mxnet/caffetranslator/Solver.java | 148 +
.../io/mxnet/caffetranslator/SolverListener.java | 58 +
.../io/mxnet/caffetranslator/SymbolGenerator.java | 29 +
.../caffetranslator/SymbolGeneratorFactory.java | 53 +
.../main/java/io/mxnet/caffetranslator/Utils.java | 42 +
.../generators/AccuracyMetricsGenerator.java | 83 +
.../caffetranslator/generators/BaseGenerator.java | 60 +
.../generators/BatchNormGenerator.java | 65 +
.../generators/ConcatGenerator.java | 49 +
.../generators/ConvolutionGenerator.java | 101 +
.../generators/DeconvolutionGenerator.java | 103 +
.../generators/DropoutGenerator.java | 43 +
.../generators/EltwiseGenerator.java | 69 +
.../caffetranslator/generators/FCGenerator.java | 82 +
.../generators/FlattenGenerator.java | 49 +
.../generators/PermuteGenerator.java | 48 +
.../generators/PluginIntLayerGenerator.java | 80 +
.../generators/PluginLayerHelper.java | 63 +
.../generators/PluginLossGenerator.java | 69 +
.../generators/PoolingGenerator.java | 86 +
.../caffetranslator/generators/PowerGenerator.java | 51 +
.../caffetranslator/generators/ReluGenerator.java | 44 +
.../caffetranslator/generators/ScaleGenerator.java | 66 +
.../generators/SoftmaxOutputGenerator.java | 43 +
.../mxnet/caffetranslator/misc/CollectStats.java | 73 +
.../mxnet/caffetranslator/misc/StatsListener.java | 103 +
.../src/main/resources/templates/accuracy.st | 20 +
.../src/main/resources/templates/activation.st | 19 +
.../src/main/resources/templates/add.st | 19 +
.../src/main/resources/templates/batchnorm.st | 32 +
.../src/main/resources/templates/concat.st | 19 +
.../src/main/resources/templates/convolution.st | 27 +
.../src/main/resources/templates/deconvolution.st | 28 +
.../src/main/resources/templates/dropout.st | 19 +
.../src/main/resources/templates/fc.st | 19 +
.../src/main/resources/templates/flatten.st | 19 +
.../src/main/resources/templates/group.st | 19 +
.../src/main/resources/templates/imports.st | 25 +
.../src/main/resources/templates/init_params.st | 25 +
.../src/main/resources/templates/iterator.st | 28 +
.../src/main/resources/templates/logging.st | 29 +
.../src/main/resources/templates/lrn.st | 19 +
.../src/main/resources/templates/lrpolicy_exp.st | 21 +
.../src/main/resources/templates/lrpolicy_inv.st | 21 +
.../main/resources/templates/lrpolicy_multistep.st | 23 +
.../src/main/resources/templates/lrpolicy_poly.st | 21 +
.../main/resources/templates/lrpolicy_sigmoid.st | 21 +
.../src/main/resources/templates/lrpolicy_step.st | 22 +
.../src/main/resources/templates/maxium.st | 19 +
.../main/resources/templates/metrics_classes.st | 104 +
.../src/main/resources/templates/mul.st | 19 +
.../src/main/resources/templates/opt_adadelta.st | 32 +
.../src/main/resources/templates/opt_adagrad.st | 28 +
.../src/main/resources/templates/opt_adam.st | 36 +
.../src/main/resources/templates/opt_nesterov.st | 28 +
.../src/main/resources/templates/opt_rmsprop.st | 32 +
.../src/main/resources/templates/opt_sgd.st | 28 +
.../src/main/resources/templates/opt_vars.st | 24 +
.../main/resources/templates/param_initializer.st | 30 +
.../src/main/resources/templates/params_loader.st | 31 +
.../src/main/resources/templates/permute.st | 19 +
.../src/main/resources/templates/pooling.st | 34 +
.../src/main/resources/templates/power.st | 19 +
.../src/main/resources/templates/runner.st | 75 +
.../src/main/resources/templates/softmaxoutput.st | 21 +
.../src/main/resources/templates/symbols.stg | 25 +
.../src/main/resources/templates/top_k_accuracy.st | 20 +
.../src/main/resources/templates/var.st | 19 +
tools/coreml/converter/_layers.py | 7 +-
tools/coreml/test/test_mxnet_converter.py | 7 +
tools/im2rec.py | 18 +-
tools/license_header.py | 6 +
1092 files changed, 69108 insertions(+), 64551 deletions(-)
rename cub => 3rdparty/cub (100%)
create mode 160000 3rdparty/googletest
create mode 160000 3rdparty/openmp
create mode 100644 cpp-package/example/feature_extract/README.md
delete mode 100644 docs/_static/jquery-1.11.1.js
create mode 100644 docs/api/python/text/text.md
create mode 100644 docs/faq/gradient_compression.md
delete mode 100644 docs/faq/torch.md
create mode 100644 docs/how_to/security.md
create mode 100644 docs/tutorials/basic/ndarray_indexing.md
delete mode 100644 docs/tutorials/general_ml/recommendation_systems.md
delete mode 100644 docs/tutorials/nlp/nce_loss.md
delete mode 100644 docs/tutorials/nlp/rnn.md
delete mode 100644 docs/tutorials/speech_recognition/baidu_warp_ctc.md
create mode 100644 docs/tutorials/speech_recognition/ctc.md
delete mode 100644 docs/tutorials/speech_recognition/speech_lstm.md
create mode 100644 example/autoencoder/README.md
create mode 100644 example/capsnet/README.md
create mode 100644 example/capsnet/capsulelayers.py
create mode 100644 example/capsnet/capsulenet.py
create mode 100644 example/capsnet/result.PNG
delete mode 100644 example/cnn_text_classification/old/text_cnn.py
create mode 100644 example/ctc/captcha_generator.py
create mode 100644 example/ctc/ctc_metrics.py
create mode 100644 example/ctc/hyperparams.py
delete mode 100644 example/ctc/lstm_ocr.py
create mode 100644 example/ctc/lstm_ocr_infer.py
create mode 100644 example/ctc/lstm_ocr_train.py
create mode 100644 example/ctc/multiproc_data.py
create mode 100644 example/ctc/ocr_iter.py
rename example/{warpctc => ctc}/sample.jpg (100%)
delete mode 100644 example/dec/dec.py
create mode 100644 example/deep-embedded-clustering/README.md
create mode 100644 example/deep-embedded-clustering/autoencoder.py
create mode 100644 example/deep-embedded-clustering/data.py
create mode 100644 example/deep-embedded-clustering/dec.py
create mode 100644 example/deep-embedded-clustering/model.py
create mode 100644 example/deep-embedded-clustering/solver.py
delete mode 100644 example/gan/dcgan.py
create mode 100755 example/gluon/word_language_model/get_wikitext2_data.sh
delete mode 100644 example/model-parallel-lstm/README.md
delete mode 100755 example/model-parallel-lstm/get_ptb_data.sh
delete mode 100644 example/model-parallel-lstm/lstm.py
delete mode 100644 example/model-parallel-lstm/lstm_ptb.py
create mode 100644 example/model-parallel/lstm/README.md
create mode 100755 example/model-parallel/lstm/get_ptb_data.sh
create mode 100644 example/model-parallel/lstm/lstm.py
create mode 100644 example/model-parallel/lstm/lstm_ptb.py
create mode 100644 example/model-parallel/matrix_factorization/README.md
create mode 100644 example/model-parallel/matrix_factorization/get_data.py
create mode 100644 example/model-parallel/matrix_factorization/model.py
create mode 100644 example/model-parallel/matrix_factorization/train.py
create mode 100644 example/module/README.md
delete mode 100644 example/module/lstm_bucketing.py
delete mode 100644 example/module/train_cifar10.ipynb
delete mode 100644 example/module/train_cifar10.py
create mode 100644 example/nce-loss/lstm_net.py
create mode 100644 example/nce-loss/random_data.py
create mode 100644 example/nce-loss/text8_data.py
create mode 100644 example/nce-loss/wordvec_net.py
create mode 100644 example/profiler/README.md
create mode 100644 example/rnn-time-major/readme.md
delete mode 100644 example/rnn/bucket_R/data_preprocessing.R
delete mode 100644 example/rnn/bucket_R/gru.cell.R
delete mode 100644 example/rnn/bucket_R/lstm.cell.R
delete mode 100644 example/rnn/bucket_R/mx.io.bucket.iter.R
delete mode 100644 example/rnn/bucket_R/rnn.R
delete mode 100644 example/rnn/bucket_R/rnn.infer.R
delete mode 100644 example/rnn/bucket_R/rnn.train.R
create mode 100644 example/rnn/bucketing/README.md
create mode 100644 example/rnn/bucketing/cudnn_lstm_bucketing.py
create mode 100755 example/rnn/bucketing/get_ptb_data.sh
create mode 100644 example/rnn/bucketing/lstm_bucketing.py
delete mode 100644 example/rnn/cudnn_lstm_bucketing.py
delete mode 100755 example/rnn/get_ptb_data.sh
delete mode 100644 example/rnn/lstm_bucketing.py
create mode 100644 example/rnn/word_lm/README.md
create mode 100644 example/rnn/word_lm/data.py
create mode 100755 example/rnn/word_lm/get_ptb_data.sh
create mode 100644 example/rnn/word_lm/model.py
create mode 100644 example/rnn/word_lm/module.py
create mode 100644 example/rnn/word_lm/train.py
create mode 100644 example/sparse/factorization_machine/README.md
create mode 100644 example/sparse/factorization_machine/metric.py
create mode 100644 example/sparse/factorization_machine/model.py
create mode 100644 example/sparse/factorization_machine/train.py
delete mode 100644 example/sparse/get_data.py
delete mode 100644 example/sparse/linear_classification.py
create mode 100644 example/sparse/linear_classification/README.md
create mode 100644 example/sparse/linear_classification/data.py
rename example/sparse/{ => linear_classification}/linear_model.py (100%)
create mode 100644 example/sparse/linear_classification/train.py
create mode 100644 example/sparse/linear_classification/weighted_softmax_ce.py
delete mode 100644 example/sparse/matrix_factorization.py
create mode 100644 example/sparse/matrix_factorization/README.md
create mode 100644 example/sparse/matrix_factorization/data.py
rename example/sparse/{matrix_fact_model.py => matrix_factorization/model.py} (100%)
create mode 100644 example/sparse/matrix_factorization/train.py
delete mode 100644 example/sparse/readme.md
delete mode 100644 example/sparse/weighted_softmax_ce.py
create mode 100644 example/sparse/wide_deep/README.md
create mode 100644 example/sparse/wide_deep/data.py
create mode 100644 example/sparse/wide_deep/model.py
create mode 100644 example/sparse/wide_deep/train.py
delete mode 100644 example/speech-demo/README.md
delete mode 100644 example/speech-demo/config_util.py
delete mode 100644 example/speech-demo/decode_mxnet.py
delete mode 100755 example/speech-demo/decode_mxnet.sh
delete mode 100644 example/speech-demo/default.cfg
delete mode 100644 example/speech-demo/default_timit.cfg
delete mode 100644 example/speech-demo/io_func/__init__.py
delete mode 100644 example/speech-demo/io_func/convert2kaldi.py
delete mode 100644 example/speech-demo/io_func/feat_io.py
delete mode 100644 example/speech-demo/io_func/feat_readers/__init__.py
delete mode 100644 example/speech-demo/io_func/feat_readers/common.py
delete mode 100644 example/speech-demo/io_func/feat_readers/reader_atrack.py
delete mode 100644 example/speech-demo/io_func/feat_readers/reader_bvec.py
delete mode 100644 example/speech-demo/io_func/feat_readers/reader_htk.py
delete mode 100644 example/speech-demo/io_func/feat_readers/reader_kaldi.py
delete mode 100644 example/speech-demo/io_func/feat_readers/stats.py
delete mode 100644 example/speech-demo/io_func/feat_readers/writer_kaldi.py
delete mode 100644 example/speech-demo/io_func/info.py
delete mode 100644 example/speech-demo/io_func/kaldi_parser.py
delete mode 100755 example/speech-demo/io_func/model_io.py
delete mode 100644 example/speech-demo/io_func/regr_feat_io.py
delete mode 100644 example/speech-demo/io_func/utils.py
delete mode 100644 example/speech-demo/io_util.py
delete mode 100644 example/speech-demo/lstm_proj.py
delete mode 100644 example/speech-demo/make_stats.py
delete mode 100644 example/speech-demo/python_wrap/Makefile
delete mode 100644 example/speech-demo/python_wrap/ctypes.cc
delete mode 100644 example/speech-demo/python_wrap/example_usage/README.txt
delete mode 100644 example/speech-demo/python_wrap/example_usage/data.ark
delete mode 100644 example/speech-demo/python_wrap/example_usage/data.scp
delete mode 100644 example/speech-demo/python_wrap/example_usage/data.txt
delete mode 100644 example/speech-demo/python_wrap/example_usage/example.py
delete mode 100755 example/speech-demo/run_ami.sh
delete mode 100755 example/speech-demo/run_timit.sh
delete mode 100644 example/speech-demo/speechSGD.py
delete mode 100644 example/speech-demo/tests/test_nothing.py
delete mode 100644 example/speech-demo/tests/test_system.py
delete mode 100644 example/speech-demo/train_lstm_proj.py
create mode 100644 example/stochastic-depth/README.md
delete mode 100644 example/torch/torch_function.py
delete mode 100644 example/torch/torch_module.py
create mode 100644 example/vae/README.md
mode change 100644 => 100755 example/vae/VAE_example.ipynb
delete mode 100644 example/warpctc/README.md
delete mode 100644 example/warpctc/infer_ocr.py
delete mode 100644 example/warpctc/lstm.py
delete mode 100644 example/warpctc/lstm_model.py
delete mode 100644 example/warpctc/lstm_ocr.py
delete mode 100644 example/warpctc/ocr_predict.py
delete mode 100644 example/warpctc/toy_ctc.py
create mode 100644 python/mxnet/contrib/text/__init__.py
create mode 100644 python/mxnet/contrib/text/_constants.py
create mode 100644 python/mxnet/contrib/text/embedding.py
create mode 100644 python/mxnet/contrib/text/glossary.py
create mode 100644 python/mxnet/contrib/text/indexer.py
create mode 100644 python/mxnet/contrib/text/utils.py
create mode 100644 python/mxnet/gluon/contrib/nn/__init__.py
create mode 100644 python/mxnet/gluon/contrib/nn/basic_layers.py
delete mode 100644 python/mxnet/gluon/model_zoo/custom_layers.py
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/AttrScope.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/Base.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/Callback.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/Context.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/DType.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/EvalMetric.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/Executor.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/ExecutorManager.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/FeedForward.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/IO.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/Initializer.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/KVStore.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/KVStoreServer.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/LRScheduler.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/LibInfo.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/Model.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/Monitor.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/NDArray.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/NameManager.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/Operator.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/Optimizer.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/Profiler.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/Random.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/RecordIO.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/Rtc.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/Serializer.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/Shape.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/Symbol.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/Visualization.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/contrib/NDArray.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/contrib/Symbol.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/io/MXDataIter.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/io/NDArrayIter.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/io/PrefetchingIter.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/io/ResizeIter.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/module/BaseModule.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/module/BucketingModule.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/module/DataParallelExecutorGroup.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/module/Module.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/module/SequentialModule.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/optimizer/AdaDelta.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/optimizer/AdaGrad.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/optimizer/Adam.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/optimizer/DCASGD.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/optimizer/NAG.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/optimizer/RMSProp.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/optimizer/SGD.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/optimizer/SGLD.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/util/NativeLibraryLoader.scala
delete mode 100644 scala-package/core/src/main/scala/ml/dmlc/mxnet/util/SerializerUtils.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/AttrScope.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/Base.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/Callback.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/Context.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/DType.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/EvalMetric.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/Executor.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/ExecutorManager.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/FeedForward.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/IO.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/Initializer.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/KVStore.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/KVStoreServer.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/LRScheduler.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/LibInfo.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/Model.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/Monitor.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/NDArray.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/NameManager.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/Operator.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/Optimizer.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/Profiler.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/Random.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/RecordIO.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/Rtc.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/Serializer.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/Shape.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/Symbol.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/Visualization.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/contrib/NDArray.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/contrib/Symbol.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/io/MXDataIter.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/io/NDArrayIter.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/io/PrefetchingIter.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/io/ResizeIter.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/module/BaseModule.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/module/BucketingModule.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/module/DataParallelExecutorGroup.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/module/Module.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/module/SequentialModule.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/optimizer/AdaDelta.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/optimizer/AdaGrad.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/optimizer/Adam.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/optimizer/DCASGD.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/optimizer/NAG.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/optimizer/RMSProp.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/optimizer/SGD.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/optimizer/SGLD.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/util/NativeLibraryLoader.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/util/SerializerUtils.scala
create mode 100644 scala-package/core/src/main/scala/org/apache/mxnet/util/WarnIfNotDisposed.scala
delete mode 100644 scala-package/core/src/test/scala/ml/dmlc/mxnet/AttrScopeSuite.scala
delete mode 100644 scala-package/core/src/test/scala/ml/dmlc/mxnet/CheckUtils.scala
delete mode 100644 scala-package/core/src/test/scala/ml/dmlc/mxnet/ExecutorSuite.scala
delete mode 100644 scala-package/core/src/test/scala/ml/dmlc/mxnet/IOSuite.scala
delete mode 100644 scala-package/core/src/test/scala/ml/dmlc/mxnet/KVStoreSuite.scala
delete mode 100644 scala-package/core/src/test/scala/ml/dmlc/mxnet/ModelParallelSuite.scala
delete mode 100644 scala-package/core/src/test/scala/ml/dmlc/mxnet/ModuleSuite.scala
delete mode 100644 scala-package/core/src/test/scala/ml/dmlc/mxnet/NDArraySuite.scala
delete mode 100644 scala-package/core/src/test/scala/ml/dmlc/mxnet/OperatorSuite.scala
delete mode 100644 scala-package/core/src/test/scala/ml/dmlc/mxnet/RandomSuite.scala
delete mode 100644 scala-package/core/src/test/scala/ml/dmlc/mxnet/RecordIOSuite.scala
delete mode 100644 scala-package/core/src/test/scala/ml/dmlc/mxnet/SerializerSuite.scala
delete mode 100644 scala-package/core/src/test/scala/ml/dmlc/mxnet/ShapeSuite.scala
delete mode 100644 scala-package/core/src/test/scala/ml/dmlc/mxnet/SymbolSuite.scala
delete mode 100644 scala-package/core/src/test/scala/ml/dmlc/mxnet/TestUtil.scala
delete mode 100644 scala-package/core/src/test/scala/ml/dmlc/mxnet/gpu/RtcSuite.scala
delete mode 100644 scala-package/core/src/test/scala/ml/dmlc/mxnet/train/ConvSuite.scala
delete mode 100644 scala-package/core/src/test/scala/ml/dmlc/mxnet/util/SerializerUtilsSuite.scala
create mode 100644 scala-package/core/src/test/scala/org/apache/mxnet/AttrScopeSuite.scala
create mode 100644 scala-package/core/src/test/scala/org/apache/mxnet/CheckUtils.scala
create mode 100644 scala-package/core/src/test/scala/org/apache/mxnet/ExecutorSuite.scala
create mode 100644 scala-package/core/src/test/scala/org/apache/mxnet/IOSuite.scala
create mode 100644 scala-package/core/src/test/scala/org/apache/mxnet/KVStoreSuite.scala
create mode 100644 scala-package/core/src/test/scala/org/apache/mxnet/ModelParallelSuite.scala
create mode 100644 scala-package/core/src/test/scala/org/apache/mxnet/ModuleSuite.scala
create mode 100644 scala-package/core/src/test/scala/org/apache/mxnet/NDArraySuite.scala
create mode 100644 scala-package/core/src/test/scala/org/apache/mxnet/OperatorSuite.scala
create mode 100644 scala-package/core/src/test/scala/org/apache/mxnet/RandomSuite.scala
create mode 100644 scala-package/core/src/test/scala/org/apache/mxnet/RecordIOSuite.scala
create mode 100644 scala-package/core/src/test/scala/org/apache/mxnet/SerializerSuite.scala
create mode 100644 scala-package/core/src/test/scala/org/apache/mxnet/ShapeSuite.scala
create mode 100644 scala-package/core/src/test/scala/org/apache/mxnet/SymbolSuite.scala
create mode 100644 scala-package/core/src/test/scala/org/apache/mxnet/TestUtil.scala
create mode 100644 scala-package/core/src/test/scala/org/apache/mxnet/gpu/RtcSuite.scala
create mode 100644 scala-package/core/src/test/scala/org/apache/mxnet/train/ConvSuite.scala
create mode 100644 scala-package/core/src/test/scala/org/apache/mxnet/util/SerializerUtilsSuite.scala
create mode 100644 scala-package/core/src/test/scala/org/apache/mxnet/util/WarnIfNotDiposedSuite.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/cnntextclassification/CNNTextClassification.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/cnntextclassification/DataHelper.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/customop/Data.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/customop/ExampleCustomOp.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/customop/ExampleCustomOpWithRtc.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/gan/GanMnist.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/gan/Module.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/gan/Viz.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/imclassification/ModelTrain.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/imclassification/TrainMnist.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/module/MnistMlp.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/module/SequentialModuleEx.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/multitask/Data.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/multitask/ExampleMultiTask.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/neuralstyle/ModelVgg19.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/neuralstyle/NeuralStyle.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/neuralstyle/end2end/Basic.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/neuralstyle/end2end/BoostInference.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/neuralstyle/end2end/BoostTrain.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/neuralstyle/end2end/DataProcessing.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/neuralstyle/end2end/GenV3.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/neuralstyle/end2end/GenV4.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/neuralstyle/end2end/ModelVgg19.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/neuralstyle/end2end/Module.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/profiler/ProfilerMatMul.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/profiler/ProfilerNDArray.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/rnn/BucketIo.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/rnn/Lstm.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/rnn/LstmBucketing.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/rnn/RnnModel.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/rnn/TestCharRnn.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/rnn/TrainCharRnn.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/rnn/Utils.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/visualization/AlexNet.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/visualization/ExampleVis.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/visualization/GoogleNet.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/visualization/Inception_BN.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/visualization/Inception_V3.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/visualization/LeNet.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/visualization/ResNet_Small.scala
delete mode 100644 scala-package/examples/src/main/scala/ml/dmlc/mxnetexamples/visualization/VGG.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/cnntextclassification/CNNTextClassification.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/cnntextclassification/DataHelper.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/customop/Data.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/customop/ExampleCustomOp.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/customop/ExampleCustomOpWithRtc.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/gan/GanMnist.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/gan/Module.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/gan/Viz.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/imclassification/ModelTrain.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/imclassification/TrainMnist.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/module/MnistMlp.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/module/SequentialModuleEx.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/multitask/Data.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/multitask/ExampleMultiTask.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/neuralstyle/ModelVgg19.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/neuralstyle/NeuralStyle.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/neuralstyle/end2end/Basic.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/neuralstyle/end2end/BoostInference.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/neuralstyle/end2end/BoostTrain.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/neuralstyle/end2end/DataProcessing.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/neuralstyle/end2end/GenV3.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/neuralstyle/end2end/GenV4.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/neuralstyle/end2end/ModelVgg19.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/neuralstyle/end2end/Module.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/profiler/ProfilerMatMul.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/profiler/ProfilerNDArray.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/rnn/BucketIo.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/rnn/Lstm.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/rnn/LstmBucketing.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/rnn/RnnModel.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/rnn/TestCharRnn.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/rnn/TrainCharRnn.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/rnn/Utils.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/visualization/AlexNet.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/visualization/ExampleVis.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/visualization/GoogleNet.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/visualization/Inception_BN.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/visualization/Inception_V3.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/visualization/LeNet.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/visualization/ResNet_Small.scala
create mode 100644 scala-package/examples/src/main/scala/org/apache/mxnetexamples/visualization/VGG.scala
delete mode 100644 scala-package/init-native/src/main/native/ml_dmlc_mxnet_init_native_c_api.cc
create mode 100644 scala-package/init-native/src/main/native/org_apache_mxnet_init_native_c_api.cc
delete mode 100644 scala-package/init/src/main/scala/ml/dmlc/mxnet/init/Base.scala
delete mode 100644 scala-package/init/src/main/scala/ml/dmlc/mxnet/init/LibInfo.scala
create mode 100644 scala-package/init/src/main/scala/org/apache/mxnet/init/Base.scala
create mode 100644 scala-package/init/src/main/scala/org/apache/mxnet/init/LibInfo.scala
delete mode 100644 scala-package/macros/src/main/scala/ml/dmlc/mxnet/NDArrayMacro.scala
delete mode 100644 scala-package/macros/src/main/scala/ml/dmlc/mxnet/SymbolMacro.scala
delete mode 100644 scala-package/macros/src/main/scala/ml/dmlc/mxnet/utils/OperatorBuildUtils.scala
create mode 100644 scala-package/macros/src/main/scala/org/apache/mxnet/NDArrayMacro.scala
create mode 100644 scala-package/macros/src/main/scala/org/apache/mxnet/SymbolMacro.scala
create mode 100644 scala-package/macros/src/main/scala/org/apache/mxnet/utils/OperatorBuildUtils.scala
delete mode 100644 scala-package/native/src/main/native/ml_dmlc_mxnet_native_c_api.cc
create mode 100644 scala-package/native/src/main/native/org_apache_mxnet_native_c_api.cc
delete mode 100644 scala-package/spark/src/main/scala/ml/dmlc/mxnet/spark/MXNDArray.scala
delete mode 100644 scala-package/spark/src/main/scala/ml/dmlc/mxnet/spark/MXNet.scala
delete mode 100644 scala-package/spark/src/main/scala/ml/dmlc/mxnet/spark/MXNetModel.scala
delete mode 100644 scala-package/spark/src/main/scala/ml/dmlc/mxnet/spark/MXNetParams.scala
delete mode 100644 scala-package/spark/src/main/scala/ml/dmlc/mxnet/spark/ParameterServer.scala
delete mode 100644 scala-package/spark/src/main/scala/ml/dmlc/mxnet/spark/example/ClassificationExample.scala
delete mode 100644 scala-package/spark/src/main/scala/ml/dmlc/mxnet/spark/io/LabeledPointIter.scala
delete mode 100644 scala-package/spark/src/main/scala/ml/dmlc/mxnet/spark/io/LongLivingDataBatch.scala
delete mode 100644 scala-package/spark/src/main/scala/ml/dmlc/mxnet/spark/io/PointIter.scala
delete mode 100644 scala-package/spark/src/main/scala/ml/dmlc/mxnet/spark/transformer/MXNet.scala
delete mode 100644 scala-package/spark/src/main/scala/ml/dmlc/mxnet/spark/utils/Img2Vector.scala
delete mode 100644 scala-package/spark/src/main/scala/ml/dmlc/mxnet/spark/utils/Network.scala
delete mode 100644 scala-package/spark/src/main/scala/ml/dmlc/mxnet/spark/utils/RepIterator.scala
create mode 100644 scala-package/spark/src/main/scala/org/apache/mxnet/spark/MXNDArray.scala
create mode 100644 scala-package/spark/src/main/scala/org/apache/mxnet/spark/MXNet.scala
create mode 100644 scala-package/spark/src/main/scala/org/apache/mxnet/spark/MXNetModel.scala
create mode 100644 scala-package/spark/src/main/scala/org/apache/mxnet/spark/MXNetParams.scala
create mode 100644 scala-package/spark/src/main/scala/org/apache/mxnet/spark/ParameterServer.scala
create mode 100644 scala-package/spark/src/main/scala/org/apache/mxnet/spark/example/ClassificationExample.scala
create mode 100644 scala-package/spark/src/main/scala/org/apache/mxnet/spark/io/LabeledPointIter.scala
create mode 100644 scala-package/spark/src/main/scala/org/apache/mxnet/spark/io/LongLivingDataBatch.scala
create mode 100644 scala-package/spark/src/main/scala/org/apache/mxnet/spark/io/PointIter.scala
create mode 100644 scala-package/spark/src/main/scala/org/apache/mxnet/spark/transformer/MXNet.scala
create mode 100644 scala-package/spark/src/main/scala/org/apache/mxnet/spark/utils/Img2Vector.scala
create mode 100644 scala-package/spark/src/main/scala/org/apache/mxnet/spark/utils/Network.scala
create mode 100644 scala-package/spark/src/main/scala/org/apache/mxnet/spark/utils/RepIterator.scala
create mode 100644 src/common/random_generator.cu
create mode 100644 src/common/random_generator.h
create mode 100644 src/kvstore/gradient_compression-inl.h
create mode 100644 src/kvstore/gradient_compression.cc
create mode 100644 src/kvstore/gradient_compression.cu
create mode 100644 src/kvstore/gradient_compression.h
create mode 100644 src/kvstore/kvstore_nccl.h
create mode 100644 src/kvstore/kvstore_utils.cc
create mode 100644 src/kvstore/kvstore_utils.cu
create mode 100644 src/kvstore/kvstore_utils.h
delete mode 100644 src/operator/activation-inl.h
delete mode 100644 src/operator/activation.cc
delete mode 100644 src/operator/activation.cu
delete mode 100644 src/operator/batch_norm-inl.h
delete mode 100644 src/operator/batch_norm.cc
delete mode 100644 src/operator/batch_norm.cu
create mode 100644 src/operator/contrib/bounding_box-inl.h
create mode 100644 src/operator/contrib/bounding_box.cc
create mode 100644 src/operator/contrib/bounding_box.cu
create mode 100644 src/operator/contrib/krprod.cc
delete mode 100644 src/operator/convolution-inl.h
delete mode 100644 src/operator/convolution.cc
delete mode 100644 src/operator/convolution.cu
delete mode 100644 src/operator/cudnn_activation-inl.h
delete mode 100644 src/operator/cudnn_algoreg-inl.h
delete mode 100644 src/operator/cudnn_batch_norm-inl.h
delete mode 100644 src/operator/cudnn_convolution-inl.h
delete mode 100644 src/operator/cudnn_deconvolution-inl.h
delete mode 100644 src/operator/cudnn_pooling-inl.h
delete mode 100644 src/operator/cudnn_softmax_activation-inl.h
delete mode 100644 src/operator/deconvolution-inl.h
delete mode 100644 src/operator/deconvolution.cc
delete mode 100644 src/operator/deconvolution.cu
delete mode 100644 src/operator/depthwise_convolution-inl.h
delete mode 100644 src/operator/depthwise_convolution_tf.cuh
delete mode 100644 src/operator/dropout-inl.h
delete mode 100644 src/operator/dropout.cc
delete mode 100644 src/operator/fully_connected-inl.h
create mode 100644 src/operator/nn/activation-inl.h
create mode 100644 src/operator/nn/activation.cc
create mode 100644 src/operator/nn/activation.cu
create mode 100644 src/operator/nn/batch_norm-inl.h
create mode 100644 src/operator/nn/batch_norm.cc
create mode 100644 src/operator/nn/batch_norm.cu
create mode 100644 src/operator/nn/convolution-inl.h
create mode 100644 src/operator/nn/convolution.cc
create mode 100644 src/operator/nn/convolution.cu
create mode 100644 src/operator/nn/cudnn/cudnn_activation-inl.h
create mode 100644 src/operator/nn/cudnn/cudnn_algoreg-inl.h
rename src/operator/{ => nn/cudnn}/cudnn_algoreg.cc (100%)
create mode 100644 src/operator/nn/cudnn/cudnn_batch_norm-inl.h
rename src/operator/{ => nn/cudnn}/cudnn_batch_norm.cc (100%)
rename src/operator/{ => nn/cudnn}/cudnn_batch_norm.cu (100%)
create mode 100644 src/operator/nn/cudnn/cudnn_convolution-inl.h
create mode 100644 src/operator/nn/cudnn/cudnn_deconvolution-inl.h
create mode 100644 src/operator/nn/cudnn/cudnn_pooling-inl.h
create mode 100644 src/operator/nn/cudnn/cudnn_softmax_activation-inl.h
create mode 100644 src/operator/nn/deconvolution-inl.h
create mode 100644 src/operator/nn/deconvolution.cc
create mode 100644 src/operator/nn/deconvolution.cu
create mode 100644 src/operator/nn/depthwise_convolution-inl.h
create mode 100644 src/operator/nn/depthwise_convolution_tf.cuh
create mode 100644 src/operator/nn/dropout-inl.h
create mode 100644 src/operator/nn/dropout.cc
rename src/operator/{ => nn}/dropout.cu (100%)
create mode 100644 src/operator/nn/fully_connected-inl.h
rename src/operator/{ => nn}/fully_connected.cc (100%)
rename src/operator/{ => nn}/fully_connected.cu (100%)
create mode 100644 src/operator/nn/pooling-inl.h
create mode 100644 src/operator/nn/pooling.cc
create mode 100644 src/operator/nn/pooling.cu
create mode 100644 src/operator/nn/softmax_activation-inl.h
create mode 100644 src/operator/nn/softmax_activation.cc
create mode 100644 src/operator/nn/softmax_activation.cu
create mode 100644 src/operator/nn/upsampling-inl.h
rename src/operator/{ => nn}/upsampling.cc (100%)
rename src/operator/{ => nn}/upsampling.cu (100%)
create mode 100644 src/operator/operator_tune-inl.h
create mode 100644 src/operator/operator_tune.cc
create mode 100644 src/operator/operator_tune.h
delete mode 100644 src/operator/pooling-inl.h
delete mode 100644 src/operator/pooling.cc
delete mode 100644 src/operator/pooling.cu
delete mode 100644 src/operator/softmax_activation-inl.h
delete mode 100644 src/operator/softmax_activation.cc
delete mode 100644 src/operator/softmax_activation.cu
create mode 100644 src/operator/tensor/square_sum.cu
delete mode 100644 src/operator/upsampling-inl.h
create mode 100644 tests/ci_build/Dockerfile.build_cuda
create mode 100644 tests/ci_build/Dockerfile.cpu_clang
create mode 100644 tests/ci_build/Dockerfile.cpu_mklml
create mode 100644 tests/ci_build/Dockerfile.gpu_mklml
delete mode 100644 tests/ci_build/Dockerfile.mklml_gpu
create mode 100755 tests/ci_build/install/ubuntu_install_nvidia.sh
create mode 100644 tests/cpp/include/test_tune.h
delete mode 100644 tests/cpp/operator/broadcast_perf.cc
create mode 100644 tests/cpp/operator/dropout_perf.cc
create mode 100644 tests/cpp/operator/tune/operator_tune_test.cc
delete mode 100644 tests/python/common/get_data.py
create mode 100644 tests/python/cpu/test_mklml.py
create mode 100644 tests/python/gpu/test_nccl.py
create mode 100644 tests/python/unittest/test_contrib_krprod.py
create mode 100644 tests/python/unittest/test_contrib_operator.py
create mode 100644 tests/python/unittest/test_contrib_text.py
create mode 100644 tools/caffe_translator/README.md
create mode 100644 tools/caffe_translator/build.gradle
create mode 100644 tools/caffe_translator/build_from_source.md
create mode 100644 tools/caffe_translator/faq.md
create mode 100644 tools/caffe_translator/gradle.properties
create mode 100644 tools/caffe_translator/gradle/wrapper/gradle-wrapper.jar
create mode 100644 tools/caffe_translator/gradle/wrapper/gradle-wrapper.properties
create mode 100755 tools/caffe_translator/gradlew
create mode 100644 tools/caffe_translator/gradlew.bat
create mode 100644 tools/caffe_translator/scripts/convert_caffe_model.py
create mode 100644 tools/caffe_translator/settings.gradle
create mode 100644 tools/caffe_translator/src/main/antlr/io/mxnet/caffetranslator/CaffePrototxt.g4
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/Config.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/Converter.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/CreateModelListener.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/GenerationHelper.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/GeneratorOutput.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/Launcher.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/Layer.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/MLModel.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/Optimizer.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/ParserHelper.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/Solver.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/SolverListener.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/SymbolGenerator.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/SymbolGeneratorFactory.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/Utils.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/generators/AccuracyMetricsGenerator.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/generators/BaseGenerator.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/generators/BatchNormGenerator.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/generators/ConcatGenerator.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/generators/ConvolutionGenerator.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/generators/DeconvolutionGenerator.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/generators/DropoutGenerator.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/generators/EltwiseGenerator.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/generators/FCGenerator.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/generators/FlattenGenerator.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/generators/PermuteGenerator.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/generators/PluginIntLayerGenerator.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/generators/PluginLayerHelper.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/generators/PluginLossGenerator.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/generators/PoolingGenerator.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/generators/PowerGenerator.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/generators/ReluGenerator.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/generators/ScaleGenerator.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/generators/SoftmaxOutputGenerator.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/misc/CollectStats.java
create mode 100644 tools/caffe_translator/src/main/java/io/mxnet/caffetranslator/misc/StatsListener.java
create mode 100644 tools/caffe_translator/src/main/resources/templates/accuracy.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/activation.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/add.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/batchnorm.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/concat.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/convolution.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/deconvolution.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/dropout.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/fc.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/flatten.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/group.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/imports.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/init_params.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/iterator.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/logging.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/lrn.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/lrpolicy_exp.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/lrpolicy_inv.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/lrpolicy_multistep.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/lrpolicy_poly.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/lrpolicy_sigmoid.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/lrpolicy_step.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/maxium.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/metrics_classes.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/mul.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/opt_adadelta.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/opt_adagrad.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/opt_adam.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/opt_nesterov.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/opt_rmsprop.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/opt_sgd.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/opt_vars.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/param_initializer.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/params_loader.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/permute.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/pooling.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/power.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/runner.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/softmaxoutput.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/symbols.stg
create mode 100644 tools/caffe_translator/src/main/resources/templates/top_k_accuracy.st
create mode 100644 tools/caffe_translator/src/main/resources/templates/var.st
--
To stop receiving notification emails like this one, please contact
['"commits@mxnet.apache.org" <co...@mxnet.apache.org>'].
[incubator-mxnet] 03/19: add stub (#8650)
Posted by jx...@apache.org.
This is an automated email from the ASF dual-hosted git repository.
jxie pushed a commit to branch vision
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git
commit c247df5010ce4f53082b0dc713d5fe175d63a57a
Author: Hu Shiwen <ya...@gmail.com>
AuthorDate: Tue Nov 14 21:28:34 2017 +0800
add stub (#8650)
* add stub
---
src/operator/image/image_random-inl.h | 79 +++++++++++++++++++++++++++++++----
1 file changed, 72 insertions(+), 7 deletions(-)
diff --git a/src/operator/image/image_random-inl.h b/src/operator/image/image_random-inl.h
index 9506607..1281a71 100644
--- a/src/operator/image/image_random-inl.h
+++ b/src/operator/image/image_random-inl.h
@@ -35,17 +35,41 @@
namespace mxnet {
namespace op {
+
+
+enum ImageRandomResource { kRandom };
+
+template<typename xpu>
+static void RandomFlip(const nnvm::NodeAttrs &attrs,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+}
+template<typename xpu>
+static void ToTensor(const nnvm::NodeAttrs &attrs,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+}
+template<typename xpu>
+static void Normalize(const nnvm::NodeAttrs &attrs,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+}
+
struct RandomBrightnessParam : public dmlc::Parameter<RandomBrightnessParam> {
float max_brightness;
DMLC_DECLARE_PARAMETER(RandomBrightnessParam) {
DMLC_DECLARE_FIELD(max_brightness)
- .set_default(0.0)
- .describe("Max Contrast.");
+ .set_default(0.0)
+ .describe("Max Brightness.");
}
};
-enum ImageRandomResource { kRandom };
-
template<typename xpu>
static void RandomBrightness(const nnvm::NodeAttrs &attrs,
const OpContext &ctx,
@@ -72,6 +96,14 @@ static void RandomBrightness(const nnvm::NodeAttrs &attrs,
}
+struct RandomContrastParam : public dmlc::Parameter<RandomContrastParam> {
+ float max_contrast;
+ DMLC_DECLARE_PARAMETER(RandomContrastParam) {
+ DMLC_DECLARE_FIELD(max_contrast)
+ .set_default(0.0)
+ .describe("Max Contrast.");
+ }
+};
/*! \brief mul_add operator */
struct mul_add {
@@ -82,7 +114,6 @@ struct mul_add {
}
};
-
template<typename xpu>
static void RandomContrast(const nnvm::NodeAttrs &attrs,
const OpContext &ctx,
@@ -99,8 +130,8 @@ static void RandomContrast(const nnvm::NodeAttrs &attrs,
Random<xpu> *prnd = ctx.requested[kRandom].get_random<xpu, real_t>(s);
- const RandomBrightnessParam ¶m = nnvm::get<RandomBrightnessParam>(attrs.parsed);
- float alpha_c = 1.0 + std::uniform_real_distribution<float>(-param.max_brightness, param.max_brightness)(prnd->GetRndEngine());
+ const RandomContrastParam ¶m = nnvm::get<RandomContrastParam>(attrs.parsed);
+ float alpha_c = 1.0 + std::uniform_real_distribution<float>(-param.max_contrast, param.max_contrast)(prnd->GetRndEngine());
const float R2YF = 0.299f;
const float G2YF = 0.587f;
@@ -129,6 +160,40 @@ static void RandomContrast(const nnvm::NodeAttrs &attrs,
}
+template<typename xpu>
+static void RandomSaturation(const nnvm::NodeAttrs &attrs,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+}
+
+template<typename xpu>
+static void RandomHue(const nnvm::NodeAttrs &attrs,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+}
+
+template<typename xpu>
+static void RandomColorJitter(const nnvm::NodeAttrs &attrs,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+}
+
+template<typename xpu>
+static void RandomLighting(const nnvm::NodeAttrs &attrs,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+}
+
+
+
} // namespace op
} // namespace mxnet
--
To stop receiving notification emails like this one, please contact
"commits@mxnet.apache.org" <co...@mxnet.apache.org>.
[incubator-mxnet] 04/19: [WIP][vision]fix (#8657)
Posted by jx...@apache.org.
This is an automated email from the ASF dual-hosted git repository.
jxie pushed a commit to branch vision
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git
commit 627ba18b527117356830c1099f9295c8641073ef
Author: Hu Shiwen <ya...@gmail.com>
AuthorDate: Wed Nov 15 12:16:23 2017 +0800
[WIP][vision]fix (#8657)
* fix
* lint
---
src/operator/image/image_random-inl.h | 118 ++++++++++++++++++++++++----------
src/operator/image/image_random.cc | 32 +++++++++
src/operator/mxnet_op.h | 6 ++
3 files changed, 122 insertions(+), 34 deletions(-)
diff --git a/src/operator/image/image_random-inl.h b/src/operator/image/image_random-inl.h
index 1281a71..fa1d6dc 100644
--- a/src/operator/image/image_random-inl.h
+++ b/src/operator/image/image_random-inl.h
@@ -31,6 +31,7 @@
#include <opencv2/core/mat.hpp>
#include "../mxnet_op.h"
#include "image_common.h"
+#include "operator/operator_common.h"
namespace mxnet {
@@ -41,32 +42,32 @@ enum ImageRandomResource { kRandom };
template<typename xpu>
static void RandomFlip(const nnvm::NodeAttrs &attrs,
- const OpContext &ctx,
- const std::vector<TBlob> &inputs,
- const std::vector<OpReqType> &req,
- const std::vector<TBlob> &outputs) {
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
}
template<typename xpu>
static void ToTensor(const nnvm::NodeAttrs &attrs,
- const OpContext &ctx,
- const std::vector<TBlob> &inputs,
- const std::vector<OpReqType> &req,
- const std::vector<TBlob> &outputs) {
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
}
template<typename xpu>
static void Normalize(const nnvm::NodeAttrs &attrs,
- const OpContext &ctx,
- const std::vector<TBlob> &inputs,
- const std::vector<OpReqType> &req,
- const std::vector<TBlob> &outputs) {
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
}
struct RandomBrightnessParam : public dmlc::Parameter<RandomBrightnessParam> {
float max_brightness;
DMLC_DECLARE_PARAMETER(RandomBrightnessParam) {
DMLC_DECLARE_FIELD(max_brightness)
- .set_default(0.0)
- .describe("Max Brightness.");
+ .set_default(0.0)
+ .describe("Max Brightness.");
}
};
@@ -88,7 +89,7 @@ static void RandomBrightness(const nnvm::NodeAttrs &attrs,
const RandomBrightnessParam ¶m = nnvm::get<RandomBrightnessParam>(attrs.parsed);
float alpha_b = 1.0 + std::uniform_real_distribution<float>(-param.max_brightness, param.max_brightness)(prnd->GetRndEngine());
MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
- MXNET_ASSIGN_REQ_SWITCH(req[0], Req,{
+ MXNET_ASSIGN_REQ_SWITCH(req[0], Req, {
mxnet_op::Kernel<mxnet_op::op_with_req<mshadow::op::mul, Req>, xpu>::Launch(
s, inputs[0].Size(), outputs[0].dptr<DType>(), inputs[0].dptr<DType>(), DType(alpha_b));
});
@@ -100,8 +101,8 @@ struct RandomContrastParam : public dmlc::Parameter<RandomContrastParam> {
float max_contrast;
DMLC_DECLARE_PARAMETER(RandomContrastParam) {
DMLC_DECLARE_FIELD(max_contrast)
- .set_default(0.0)
- .describe("Max Contrast.");
+ .set_default(0.0)
+ .describe("Max Contrast.");
}
};
@@ -139,7 +140,7 @@ static void RandomContrast(const nnvm::NodeAttrs &attrs,
static const float coeffs0[] = { R2YF, G2YF, B2YF };
MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
- auto input_3d = input.FlatTo3D<xpu, DType>(s);
+ auto input_3d = input.get<xpu, 3, DType>(s);
DType sum = (DType)0.0;
for (int c = 0; c < channel; ++c) {
for (int h = 0; h < hight; ++h) {
@@ -160,36 +161,85 @@ static void RandomContrast(const nnvm::NodeAttrs &attrs,
}
+struct RandomSaturationParam : public dmlc::Parameter<RandomSaturationParam> {
+ float max_saturation;
+ DMLC_DECLARE_PARAMETER(RandomSaturationParam) {
+ DMLC_DECLARE_FIELD(max_saturation)
+ .set_default(0.0)
+ .describe("Max Saturation.");
+ }
+};
+
template<typename xpu>
static void RandomSaturation(const nnvm::NodeAttrs &attrs,
- const OpContext &ctx,
- const std::vector<TBlob> &inputs,
- const std::vector<OpReqType> &req,
- const std::vector<TBlob> &outputs) {
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+ using namespace mshadow;
+ auto input = inputs[0];
+ auto output = outputs[0];
+ int channel = input.shape_[0];
+ int hight = input.shape_[1];
+ int weight = input.shape_[2];
+ Stream<xpu> *s = ctx.get_stream<xpu>();
+ Random<xpu> *prnd = ctx.requested[kRandom].get_random<xpu, real_t>(s);
+ const RandomSaturationParam ¶m = nnvm::get<RandomSaturationParam>(attrs.parsed);
+ float alpha_s = 1.0 + std::uniform_real_distribution<float>(-param.max_saturation, param.max_saturation)(prnd->GetRndEngine());
+ float alpha_o = 1 - alpha_s;
+ const float R2YF = 0.299f;
+ const float G2YF = 0.587f;
+ const float B2YF = 0.114f;
+ static const float coeffs0[] = { R2YF, G2YF, B2YF };
+
+
+ MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
+ MXNET_ASSIGN_REQ_SWITCH(req[0], Req, {
+ auto input_3d = input.get<xpu, 3, DType>(s);
+ auto output_3d = output.get<xpu, 3, DType>(s);
+ switch (channel) {
+ case 1:
+ Assign(output_3d, Req, input_3d)
+ break;
+ case 3:
+ for (int h = 0; h < hight; ++h) {
+ for (int w = 0; w < weight; ++w) {
+ float gray = input_3d[0][h][w] * R2YF + input_3d[1][h][w] * G2YF + input_3d[2][h][w] * B2YF;
+ Assign(output_3d[0][h][w], Req, DType(gray * alpha_s + input_3d[0][h][w] * alpha_o))
+ }
+ }
+ break;
+ default:
+ LOG(FATAL) << "not support channel" << channel;
+
+ }
+ });
+ });
+
}
template<typename xpu>
static void RandomHue(const nnvm::NodeAttrs &attrs,
- const OpContext &ctx,
- const std::vector<TBlob> &inputs,
- const std::vector<OpReqType> &req,
- const std::vector<TBlob> &outputs) {
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
}
template<typename xpu>
static void RandomColorJitter(const nnvm::NodeAttrs &attrs,
- const OpContext &ctx,
- const std::vector<TBlob> &inputs,
- const std::vector<OpReqType> &req,
- const std::vector<TBlob> &outputs) {
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
}
template<typename xpu>
static void RandomLighting(const nnvm::NodeAttrs &attrs,
- const OpContext &ctx,
- const std::vector<TBlob> &inputs,
- const std::vector<OpReqType> &req,
- const std::vector<TBlob> &outputs) {
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
}
diff --git a/src/operator/image/image_random.cc b/src/operator/image/image_random.cc
index ade43d8..32648bb 100644
--- a/src/operator/image/image_random.cc
+++ b/src/operator/image/image_random.cc
@@ -49,5 +49,37 @@ NNVM_REGISTER_OP(_image_random_brightness)
.add_argument("data", "NDArray-or-Symbol", "The input.")
.add_arguments(RandomBrightnessParam::__FIELDS__());
+DMLC_REGISTER_PARAMETER(RandomContrastParam);
+NNVM_REGISTER_OP(_image_random_contrast)
+.describe(R"code()code" ADD_FILELINE)
+.set_num_inputs(1)
+.set_num_outputs(1)
+.set_attr_parser(ParamParser<RandomContrastParam>)
+.set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& attrs) {
+ return std::vector<ResourceRequest>{ResourceRequest::kRandom};
+})
+.set_attr<nnvm::FInferShape>("FInferShape", ElemwiseShape<1, 1>)
+.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
+.set_attr<FCompute>("FCompute<cpu>", RandomContrast<cpu>)
+.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
+.add_argument("data", "NDArray-or-Symbol", "The input.")
+.add_arguments(RandomContrastParam::__FIELDS__());
+
+DMLC_REGISTER_PARAMETER(RandomSaturationParam);
+NNVM_REGISTER_OP(_image_random_saturation)
+.describe(R"code()code" ADD_FILELINE)
+.set_num_inputs(1)
+.set_num_outputs(1)
+.set_attr_parser(ParamParser<RandomSaturationParam>)
+.set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& attrs) {
+ return std::vector<ResourceRequest>{ResourceRequest::kRandom};
+})
+.set_attr<nnvm::FInferShape>("FInferShape", ElemwiseShape<1, 1>)
+.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
+.set_attr<FCompute>("FCompute<cpu>", RandomSaturation<cpu>)
+.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
+.add_argument("data", "NDArray-or-Symbol", "The input.")
+.add_arguments(RandomSaturationParam::__FIELDS__());
+
}
}
diff --git a/src/operator/mxnet_op.h b/src/operator/mxnet_op.h
index 081e40a..43f3a55 100644
--- a/src/operator/mxnet_op.h
+++ b/src/operator/mxnet_op.h
@@ -366,6 +366,12 @@ struct op_with_req {
KERNEL_ASSIGN(out[i], req, OP::Map(in[i], value));
}
+ /*! \brief input is tensor and two scalar value */
+ template<typename DType>
+ MSHADOW_XINLINE static void Map(int i, DType *out, const DType *in, const DType value_1, const DType value_2) {
+ KERNEL_ASSIGN(out[i], req, OP::Map(in[i], value_1, value_2));
+ }
+
/*! \brief No inputs (ie fill to constant value) */
template<typename DType>
MSHADOW_XINLINE static void Map(int i, DType *out) {
--
To stop receiving notification emails like this one, please contact
"commits@mxnet.apache.org" <co...@mxnet.apache.org>.
[incubator-mxnet] 16/19: Vision (#8856)
Posted by jx...@apache.org.
This is an automated email from the ASF dual-hosted git repository.
jxie pushed a commit to branch vision
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git
commit d07f3afdd42282803dfc53fc9016a343418f896f
Author: Eric Junyuan Xie <pi...@users.noreply.github.com>
AuthorDate: Tue Nov 28 17:14:26 2017 -0800
Vision (#8856)
* refactor
* fix
* fix
---
src/operator/image/image_random-inl.h | 756 ++++++++++++++++++----------------
src/operator/image/image_random.cc | 84 ++--
2 files changed, 455 insertions(+), 385 deletions(-)
diff --git a/src/operator/image/image_random-inl.h b/src/operator/image/image_random-inl.h
index 3bee843..9d10a30 100644
--- a/src/operator/image/image_random-inl.h
+++ b/src/operator/image/image_random-inl.h
@@ -38,18 +38,19 @@
namespace mxnet {
namespace op {
+namespace image {
-inline bool CheckIsImage(const TBlob &image) {
- CHECK_EQ(image.type_flag_, mshadow::kUint8) << "input type is not an image.";
- CHECK_EQ(image.ndim(), 3) << "input dimension is not 3.";
- CHECK(image.shape_[2] == 1 || image.shape_[2] == 3) << "image channel should be 1 or 3.";
-}
-
-static void RandomFlip(const nnvm::NodeAttrs &attrs,
- const OpContext &ctx,
- const std::vector<TBlob> &inputs,
- const std::vector<OpReqType> &req,
- const std::vector<TBlob> &outputs) {
+inline bool ToTensorShape(const nnvm::NodeAttrs& attrs,
+ std::vector<TShape> *in_attrs,
+ std::vector<TShape> *out_attrs) {
+ CHECK_EQ(in_attrs->size(), 1U);
+ CHECK_EQ(out_attrs->size(), 1U);
+ TShape &shp = (*in_attrs)[0];
+ if (!shp.ndim()) return false;
+ CHECK_EQ(shp.ndim(), 3)
+ << "Input image must have shape (height, width, channels), but got " << shp;
+ SHAPE_ASSIGN_CHECK(*out_attrs, 0, TShape({shp[2], shp[0], shp[1]}));
+ return true;
}
inline bool ToTensorType(const nnvm::NodeAttrs& attrs,
@@ -57,47 +58,39 @@ inline bool ToTensorType(const nnvm::NodeAttrs& attrs,
std::vector<int> *out_attrs) {
CHECK_EQ(in_attrs->size(), 1U);
CHECK_EQ(out_attrs->size(), 1U);
- CHECK_EQ((*in_attrs)[0], mshadow::kUint8)
- << "`to_tensor` only supports uint8 input";
TYPE_ASSIGN_CHECK(*out_attrs, 0, mshadow::kFloat32);
return (*in_attrs)[0] != -1;
}
-inline bool ToTensorShape(const nnvm::NodeAttrs& attrs,
- std::vector<TShape> *in_attrs,
- std::vector<TShape> *out_attrs) {
- CHECK_EQ(in_attrs->size(), 1U);
- CHECK_EQ(out_attrs->size(), 1U);
- TShape &shp = (*in_attrs)[0];
- CHECK_EQ(shp.ndim(), 3U) << "`to_tensor` only supports 3 dimensions";
- TShape ret(3);
- ret[0] = shp[2];
- ret[1] = shp[0];
- ret[2] = shp[1];
- SHAPE_ASSIGN_CHECK(*out_attrs, 0, ret);
- return true;
-}
-
-static void ToTensor(const nnvm::NodeAttrs &attrs,
+void ToTensor(const nnvm::NodeAttrs &attrs,
const OpContext &ctx,
const std::vector<TBlob> &inputs,
const std::vector<OpReqType> &req,
const std::vector<TBlob> &outputs) {
CHECK_EQ(req[0], kWriteTo)
<< "`to_tensor` does not support inplace";
- CheckIsImage(inputs[0]);
int length = inputs[0].shape_[0] * inputs[0].shape_[1];
int channel = inputs[0].shape_[2];
- float* output = outputs[0].dptr<float>();
- uint8_t* input = inputs[0].dptr<uint8_t>();
+ MSHADOW_TYPE_SWITCH(inputs[0].type_flag_, DType, {
+ float* output = outputs[0].dptr<float>();
+ DType* input = inputs[0].dptr<DType>();
- for (int l = 0; l < length; ++l) {
- for (int c = 0; c < channel; ++c) {
- output[c*length + l] = static_cast<float>(input[l*channel + c]) / 255.0f;
+ for (int l = 0; l < length; ++l) {
+ for (int c = 0; c < channel; ++c) {
+ output[c*length + l] = static_cast<float>(input[l*channel + c]) / 255.0f;
+ }
}
- }
+ });
+}
+
+inline bool TensorShape(const nnvm::NodeAttrs& attrs,
+ std::vector<TShape> *in_attrs,
+ std::vector<TShape> *out_attrs) {
+ TShape& dshape = (*in_attrs)[0];
+ SHAPE_ASSIGN_CHECK(*out_attrs, 0, dshape);
+ return true;
}
struct NormalizeParam : public dmlc::Parameter<NormalizeParam> {
@@ -117,20 +110,28 @@ inline bool NormalizeShape(const nnvm::NodeAttrs& attrs,
const NormalizeParam ¶m = nnvm::get<NormalizeParam>(attrs.parsed);
const auto& dshape = (*in_attrs)[0];
if (!dshape.ndim()) return false;
- CHECK_EQ(dshape.ndim(), 3)
- << "Input must have 3 dimensions";
+ CHECK_EQ(dshape.ndim(), 3)
+ << "Input tensor must have shape (channels, height, width), but got "
+ << dshape;
auto nchannels = dshape[0];
+ CHECK(nchannels == 3 || nchannels == 1)
+ << "The first dimension of input tensor must be the channel dimension with "
+ << "either 1 or 3 elements, but got input with shape " << dshape;
CHECK(param.mean.ndim() == 1 || param.mean.ndim() == nchannels)
- << "mean must have either 1 or " << nchannels << " elements";
+ << "Invalid mean for input with shape " << dshape
+ << ". mean must have either 1 or " << nchannels
+ << " elements, but got " << param.mean;
CHECK(param.std.ndim() == 1 || param.std.ndim() == nchannels)
- << "std must have either 1 or " << nchannels << " elements";
+ << "Invalid std for input with shape " << dshape
+ << ". std must have either 1 or " << nchannels
+ << " elements, but got " << param.std;
SHAPE_ASSIGN_CHECK(*out_attrs, 0, dshape);
+ return true;
}
-
-static void Normalize(const nnvm::NodeAttrs &attrs,
+void Normalize(const nnvm::NodeAttrs &attrs,
const OpContext &ctx,
const std::vector<TBlob> &inputs,
const std::vector<OpReqType> &req,
@@ -154,214 +155,241 @@ static void Normalize(const nnvm::NodeAttrs &attrs,
});
}
-struct FlipParam : public dmlc::Parameter<FlipParam> {
- int axis;
- DMLC_DECLARE_PARAMETER(FlipParam) {
- DMLC_DECLARE_FIELD(axis)
- .describe("0 or 1. 0 for horizontal flip, 1 for vertical flip.");
- }
-};
+template<typename DType>
+inline DType saturate_cast(const float& src) {
+ return static_cast<DType>(src);
+}
-#define SWAP_IF_INPLACE(dst, dst_idx, src, src_idx) \
- if (dst == src) { \
- std::swap(dst[dst_idx], src[src_idx]); \
- } else { \
- dst[dst_idx] = src[src_idx]; \
- }
+template<>
+inline uint8_t saturate_cast(const float& src) {
+ return std::min(std::max(src, 0.f), 255.f);
+}
+
+inline bool ImageShape(const nnvm::NodeAttrs& attrs,
+ std::vector<TShape> *in_attrs,
+ std::vector<TShape> *out_attrs) {
+ TShape& dshape = (*in_attrs)[0];
+ CHECK_EQ(dshape.ndim(), 3)
+ << "Input image must have shape (height, width, channels), but got " << dshape;
+ auto nchannels = dshape[dshape.ndim()-1];
+ CHECK(nchannels == 3 || nchannels == 1)
+ << "The last dimension of input image must be the channel dimension with "
+ << "either 1 or 3 elements, but got input with shape " << dshape;
+ SHAPE_ASSIGN_CHECK(*out_attrs, 0, dshape);
+ return true;
+}
template<typename DType>
-static void FlipImpl(const TShape &shape, DType *src, DType *dst, int axis) {
- const int height = shape[0];
- const int width = shape[1];
- const int nchannel = shape[2];
-
- const int length = width * nchannel;
- const int height_stride = (src == dst && axis == 1) ? (height >> 1) : height;
- const int width_stride = (src == dst && axis == 0) ? (width >> 1) : width;
-
- for (int h = 0; h < height_stride; ++h) {
- const int h_dst = (axis == 0) ? h : (height - h);
- for (int w = 0; w < width_stride; ++w) {
- const int w_dst = (axis == 0) ? (width - w) : w;
- const int idx_dst = h_dst * length + w_dst * nchannel;
- const int idx_src = h * length + w * nchannel;
- SWAP_IF_INPLACE(dst, idx_dst, src, idx_src);
- if (nchannel > 1) {
- SWAP_IF_INPLACE(dst, idx_dst + 1, src, idx_src + 1);
- SWAP_IF_INPLACE(dst, idx_dst + 2, src, idx_src + 2);
+void FlipImpl(const TShape &shape, DType *src, DType *dst, int axis) {
+ int head = 1, mid = shape[axis], tail = 1;
+ for (int i = 0; i < axis; ++i) head *= shape[i];
+ for (int i = axis+1; i < shape.ndim(); ++i) tail *= shape[i];
+
+ for (int i = 0; i < head; ++i) {
+ for (int j = 0; j < (mid >>2); ++j) {
+ int idx1 = (i*mid + j)*tail;
+ int idx2 = idx1 + (mid - (j<<2))*tail;
+ for (int k = 0; k < tail; ++k, ++idx1, ++idx2) {
+ DType tmp = src[idx1];
+ dst[idx1] = src[idx2];
+ dst[idx2] = tmp;
}
}
}
}
-static void Flip(const nnvm::NodeAttrs &attrs,
- const OpContext &ctx,
- const std::vector<TBlob> &inputs,
- const std::vector<OpReqType> &req,
- const std::vector<TBlob> &outputs) {
- const FlipParam ¶m = nnvm::get<FlipParam>(attrs.parsed);
- CHECK(param.axis == 0 || param.axis == 1) << "flip axis must be 0 or 1.";
- CheckIsImage(inputs[0]);
- const TShape& ishape = inputs[0].shape_;
+void RandomHorizontalFlip(
+ const nnvm::NodeAttrs &attrs,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+ using namespace mshadow;
+ Stream<cpu> *s = ctx.get_stream<cpu>();
+ Random<cpu> *prnd = ctx.requested[0].get_random<cpu, float>(s);
+ if (std::bernoulli_distribution()(prnd->GetRndEngine())) return;
+ MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
+ FlipImpl(inputs[0].shape_, inputs[0].dptr<DType>(),
+ outputs[0].dptr<DType>(), 1);
+ });
+}
+
+void RandomVerticalFlip(
+ const nnvm::NodeAttrs &attrs,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+ using namespace mshadow;
+ Stream<cpu> *s = ctx.get_stream<cpu>();
+ Random<cpu> *prnd = ctx.requested[0].get_random<cpu, float>(s);
+ if (std::bernoulli_distribution()(prnd->GetRndEngine())) return;
MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
- FlipImpl(ishape, inputs[0].dptr<DType>(), outputs[0].dptr<DType>(), param.axis);
+ FlipImpl(inputs[0].shape_, inputs[0].dptr<DType>(),
+ outputs[0].dptr<DType>(), 0);
});
}
-struct RandomBrightnessParam : public dmlc::Parameter<RandomBrightnessParam> {
- float max_brightness;
- DMLC_DECLARE_PARAMETER(RandomBrightnessParam) {
- DMLC_DECLARE_FIELD(max_brightness)
+struct RandomEnhanceParam : public dmlc::Parameter<RandomEnhanceParam> {
+ float min_factor;
+ float max_factor;
+ DMLC_DECLARE_PARAMETER(RandomEnhanceParam) {
+ DMLC_DECLARE_FIELD(min_factor)
+ .set_lower_bound(0.0)
+ .describe("Minimum factor.");
+ DMLC_DECLARE_FIELD(max_factor)
.set_lower_bound(0.0)
- .describe("Max Brightness.");
+ .describe("Maximum factor.");
}
};
-static void RandomBrightness(const nnvm::NodeAttrs &attrs,
- const OpContext &ctx,
- const std::vector<TBlob> &inputs,
- const std::vector<OpReqType> &req,
- const std::vector<TBlob> &outputs) {
+inline void AdjustBrightnessImpl(const float& alpha_b,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
using namespace mshadow;
- const RandomBrightnessParam ¶m = nnvm::get<RandomBrightnessParam>(attrs.parsed);
-
int length = inputs[0].Size();
- uint8_t* output = outputs[0].dptr<uint8_t>();
- uint8_t* input = inputs[0].dptr<uint8_t>();
+ MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
+ DType* output = outputs[0].dptr<DType>();
+ DType* input = inputs[0].dptr<DType>();
+ for (int l = 0; l < length; ++l) {
+ float val = static_cast<float>(input[l]) * alpha_b;
+ output[l] = saturate_cast<DType>(val);
+ }
+ });
+}
+
+void RandomBrightness(const nnvm::NodeAttrs &attrs,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+ using namespace mshadow;
+ const RandomEnhanceParam ¶m = nnvm::get<RandomEnhanceParam>(attrs.parsed);
+
Stream<cpu> *s = ctx.get_stream<cpu>();
Random<cpu> *prnd = ctx.requested[0].get_random<cpu, float>(s);
- float alpha_b = 1.0 + std::uniform_real_distribution<float>(
- -param.max_brightness, param.max_brightness)(prnd->GetRndEngine());
+ float alpha_b = std::uniform_real_distribution<float>(
+ param.min_factor, param.max_factor)(prnd->GetRndEngine());
- for (int l = 0; l < length; ++l) {
- float val = static_cast<float>(input[l]) * alpha_b;
- val = std::min(std::max(val, 0.f), 255.f);
- output[l] = static_cast<uint8_t>(val);
- }
+ AdjustBrightnessImpl(alpha_b, ctx, inputs, req, outputs);
}
+inline void AdjustContrastImpl(const float& alpha_c,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+ using namespace mshadow;
+ static const float coef[] = { 0.299f, 0.587f, 0.114f };
-struct RandomContrastParam : public dmlc::Parameter<RandomContrastParam> {
- float max_contrast;
- DMLC_DECLARE_PARAMETER(RandomContrastParam) {
- DMLC_DECLARE_FIELD(max_contrast)
- .set_lower_bound(0.0)
- .describe("Max Contrast.");
- }
-};
+ int length = inputs[0].shape_[0] * inputs[0].shape_[1];
+ int nchannels = inputs[0].shape_[2];
+ MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
+ DType* output = outputs[0].dptr<DType>();
+ DType* input = inputs[0].dptr<DType>();
+
+ float sum = 0.f;
+ if (nchannels > 1) {
+ for (int l = 0; l < length; ++l) {
+ for (int c = 0; c < 3; ++c) sum += input[l*3 + c] * coef[c];
+ }
+ } else {
+ for (int l = 0; l < length; ++l) sum += input[l];
+ }
+ float gray_mean = sum / static_cast<float>(length);
+ float beta = (1 - alpha_c) * gray_mean;
-static void RandomContrast(const nnvm::NodeAttrs &attrs,
+ for (int l = 0; l < length * nchannels; ++l) {
+ float val = input[l] * alpha_c + beta;
+ output[l] = saturate_cast<DType>(val);
+ }
+ });
+}
+
+inline void RandomContrast(const nnvm::NodeAttrs &attrs,
const OpContext &ctx,
const std::vector<TBlob> &inputs,
const std::vector<OpReqType> &req,
const std::vector<TBlob> &outputs) {
using namespace mshadow;
+ const RandomEnhanceParam ¶m = nnvm::get<RandomEnhanceParam>(attrs.parsed);
+
+
+ Stream<cpu> *s = ctx.get_stream<cpu>();
+ Random<cpu> *prnd = ctx.requested[0].get_random<cpu, real_t>(s);
+ float alpha_c = std::uniform_real_distribution<float>(
+ param.min_factor, param.max_factor)(prnd->GetRndEngine());
+
+ AdjustContrastImpl(alpha_c, ctx, inputs, req, outputs);
+}
+
+inline void AdjustSaturationImpl(const float& alpha_s,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
static const float coef[] = { 0.299f, 0.587f, 0.114f };
- const RandomContrastParam ¶m = nnvm::get<RandomContrastParam>(attrs.parsed);
int length = inputs[0].shape_[0] * inputs[0].shape_[1];
int nchannels = inputs[0].shape_[2];
- uint8_t* output = outputs[0].dptr<uint8_t>();
- uint8_t* input = inputs[0].dptr<uint8_t>();
+ float alpha_o = 1.f - alpha_s;
- Stream<cpu> *s = ctx.get_stream<cpu>();
- Random<cpu> *prnd = ctx.requested[0].get_random<cpu, real_t>(s);
- float alpha_c = 1.0 + std::uniform_real_distribution<float>(
- -param.max_contrast, param.max_contrast)(prnd->GetRndEngine());
+ MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
+ DType* output = outputs[0].dptr<DType>();
+ DType* input = inputs[0].dptr<DType>();
- float sum = 0.f;
- if (nchannels > 1) {
- for (int l = 0; l < length; ++l) {
- for (int c = 0; c < nchannels; ++c) sum += input[l*nchannels + c] * coef[c];
+ if (nchannels == 1) {
+ for (int l = 0; l < length; ++l) output[l] = input[l];
+ return;
}
- } else {
- for (int l = 0; l < length; ++l) sum += input[l];
- }
- float gray_mean = sum / static_cast<float>(length);
- float beta = (1 - alpha_c) * gray_mean;
- for (int l = 0; l < length * nchannels; ++l) {
- float val = input[l] * alpha_c + beta;
- val = std::min(std::max(val, 0.f), 255.f);
- output[l] = static_cast<uint8_t>(val);
- }
+ for (int l = 0; l < length; ++l) {
+ float gray = 0.f;
+ for (int c = 0; c < 3; ++c) {
+ gray = input[l*3 + c] * coef[c];
+ }
+ gray *= alpha_o;
+ for (int c = 0; c < 3; ++c) {
+ float val = gray + input[l*3 + c] * alpha_s;
+ output[l*3 + c] = saturate_cast<DType>(val);
+ }
+ }
+ });
}
-struct RandomSaturationParam : public dmlc::Parameter<RandomSaturationParam> {
- float max_saturation;
- DMLC_DECLARE_PARAMETER(RandomSaturationParam) {
- DMLC_DECLARE_FIELD(max_saturation)
- .set_default(0.0)
- .describe("Max Saturation.");
- }
-};
-
-static void RandomSaturation(const nnvm::NodeAttrs &attrs,
+inline void RandomSaturation(const nnvm::NodeAttrs &attrs,
const OpContext &ctx,
const std::vector<TBlob> &inputs,
const std::vector<OpReqType> &req,
const std::vector<TBlob> &outputs) {
using namespace mshadow;
- const RandomSaturationParam ¶m = nnvm::get<RandomSaturationParam>(attrs.parsed);
- static const float coef[] = { 0.299f, 0.587f, 0.114f };
-
- int length = inputs[0].shape_[0] * inputs[0].shape_[1];
- int nchannels = inputs[0].shape_[2];
-
- uint8_t* output = outputs[0].dptr<uint8_t>();
- uint8_t* input = inputs[0].dptr<uint8_t>();
+ const RandomEnhanceParam ¶m = nnvm::get<RandomEnhanceParam>(attrs.parsed);
Stream<cpu> *s = ctx.get_stream<cpu>();
Random<cpu> *prnd = ctx.requested[0].get_random<cpu, real_t>(s);
- float alpha_s = 1.f + std::uniform_real_distribution<float>(
- -param.max_saturation, param.max_saturation)(prnd->GetRndEngine());
- float alpha_o = 1.f - alpha_s;
-
- if (nchannels == 1) {
- for (int l = 0; l < length * nchannels; ++l) output[l] = input[l];
- return;
- }
+ float alpha_s = std::uniform_real_distribution<float>(
+ param.min_factor, param.max_factor)(prnd->GetRndEngine());
- for (int l = 0; l < length; ++l) {
- float gray = 0.f;
- for (int c = 0; c < nchannels; ++c) {
- gray = input[l*nchannels + c] * coef[c];
- }
- gray *= alpha_o;
- for (int c = 0; c < nchannels; ++c) {
- float val = gray + input[l*nchannels + c] * alpha_s;
- val = std::min(std::max(val, 0.f), 255.f);
- output[l*nchannels + c] = static_cast<uint8_t>(val);
- }
- }
+ AdjustSaturationImpl(alpha_s, ctx, inputs, req, outputs);
}
-struct RandomHueParam : public dmlc::Parameter<RandomHueParam> {
- float max_hue;
- DMLC_DECLARE_PARAMETER(RandomHueParam) {
- DMLC_DECLARE_FIELD(max_hue)
- .set_default(0.0)
- .describe("Max Hue.");
- }
-};
-
-template <typename DType> static
-void RGB2HLSConvert(const DType src_r,
- const DType src_g,
- const DType src_b,
- DType *dst_h,
- DType *dst_l,
- DType *dst_s
- ) {
- DType b = src_b, g = src_g, r = src_r;
- DType h = 0.f, s = 0.f, l;
- DType vmin;
- DType vmax;
- DType diff;
+void RGB2HLSConvert(const float& src_r,
+ const float& src_g,
+ const float& src_b,
+ float *dst_h,
+ float *dst_l,
+ float *dst_s) {
+ float b = src_b / 255.f, g = src_g / 255.f, r = src_r / 255.f;
+ float h = 0.f, s = 0.f, l;
+ float vmin;
+ float vmax;
+ float diff;
vmax = vmin = r;
vmax = fmax(vmax, g);
@@ -372,7 +400,7 @@ void RGB2HLSConvert(const DType src_r,
diff = vmax - vmin;
l = (vmax + vmin) * 0.5f;
- if (diff > std::numeric_limits<DType>::epsilon()) {
+ if (diff > std::numeric_limits<float>::epsilon()) {
s = (l < 0.5f) * diff / (vmax + vmin);
s += (l >= 0.5f) * diff / (2.0f - vmax - vmin);
@@ -389,23 +417,20 @@ void RGB2HLSConvert(const DType src_r,
*dst_s = s;
}
-
-static int c_HlsSectorData[6][3] = {
- { 1, 3, 0 },
- { 1, 0, 2 },
- { 3, 0, 1 },
- { 0, 2, 1 },
- { 0, 1, 3 },
- { 2, 1, 0 }
-};
-
-template <typename DType> static void HLS2RGBConvert(const DType src_h,
- const DType src_l,
- const DType src_s,
- DType *dst_r,
- DType *dst_g,
- DType *dst_b) {
-
+void HLS2RGBConvert(const float& src_h,
+ const float& src_l,
+ const float& src_s,
+ float *dst_r,
+ float *dst_g,
+ float *dst_b) {
+ static const int c_HlsSectorData[6][3] = {
+ { 1, 3, 0 },
+ { 1, 0, 2 },
+ { 3, 0, 1 },
+ { 0, 2, 1 },
+ { 0, 1, 3 },
+ { 2, 1, 0 }
+ };
float h = src_h, l = src_l, s = src_s;
float b = l, g = l, r = l;
@@ -415,6 +440,8 @@ template <typename DType> static void HLS2RGBConvert(const DType src_h,
p2 += (l > 0.5f) * (l + s - l * s);
float p1 = 2 * l - p2;
+ h *= 1.f / 60.f;
+
if (h < 0) {
do { h += 6; } while (h < 0);
} else if (h >= 6) {
@@ -436,177 +463,202 @@ template <typename DType> static void HLS2RGBConvert(const DType src_h,
r = tab[c_HlsSectorData[sector][2]];
}
- *dst_b = b;
- *dst_g = g;
- *dst_r = r;
+ *dst_b = b * 255.f;
+ *dst_g = g * 255.f;
+ *dst_r = r * 255.f;
}
-template<typename xpu, typename DType>
-static void RandomHueKernal(const TBlob &input,
- const TBlob &output,
- Stream<xpu> *s,
- int hight,
- int weight,
- DType alpha) {
- auto input_3d = input.get<xpu, 3, DType>(s);
- auto output_3d = output.get<xpu, 3, DType>(s);
- for (int h_index = 0; h_index < hight; ++h_index) {
- for (int w_index = 0; w_index < weight; ++w_index) {
- DType h;
- DType l;
- DType s;
- RGB2HLSConvert(input_3d[0][h_index][w_index],
- input_3d[1][h_index][w_index],
- input_3d[2][h_index][w_index],
- &h, &l, &s);
- h += alpha;
- h = std::max(DType(0), std::min(DType(180), h));
-
- HLS2RGBConvert(
- h, l, s,
- &output_3d[0][h_index][w_index],
- &output_3d[1][h_index][w_index],
- &output_3d[2][h_index][w_index]);
+void AdjustHueImpl(float alpha,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+ int length = inputs[0].shape_[0] * inputs[0].shape_[1];
+ if (inputs[0].shape_[2] == 1) return;
+
+ MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
+ DType* input = inputs[0].dptr<DType>();
+ DType* output = outputs[0].dptr<DType>();
+
+ for (int i = 0; i < length; ++i) {
+ float h, l, s;
+ float r = static_cast<float>(*(input++));
+ float g = static_cast<float>(*(input++));
+ float b = static_cast<float>(*(input++));
+ RGB2HLSConvert(r, g, b, &h, &l, &s);
+ h += alpha * 360.f;
+ HLS2RGBConvert(h, l, s, &r, &g, &b);
+ *(output++) = saturate_cast<DType>(r);
+ *(output++) = saturate_cast<DType>(g);
+ *(output++) = saturate_cast<DType>(b);
}
- }
+ });
}
-template<typename xpu>
-static void RandomHue(const nnvm::NodeAttrs &attrs,
- const OpContext &ctx,
- const std::vector<TBlob> &inputs,
- const std::vector<OpReqType> &req,
- const std::vector<TBlob> &outputs) {
+void RandomHue(const nnvm::NodeAttrs &attrs,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
using namespace mshadow;
- auto input = inputs[0];
- auto output = outputs[0];
- int channel = input.shape_[0];
- int hight = input.shape_[1];
- int weight = input.shape_[2];
- Stream<xpu> *s = ctx.get_stream<xpu>();
- Random<xpu> *prnd = ctx.requested[kRandom].get_random<xpu, real_t>(s);
-
- const RandomHueParam ¶m = nnvm::get<RandomHueParam>(attrs.parsed);
- float alpha = std::uniform_real_distribution<float>(
- -param.max_hue, param.max_hue)(prnd->GetRndEngine());
- auto output_float = output.get<xpu, 3, float>(s);
-
- MSHADOW_TYPE_SWITCH(input.type_flag_, DType, {
- RandomHueKernal<xpu, DType>(input, output, s, hight, weight, alpha);
- });
+ const RandomEnhanceParam ¶m = nnvm::get<RandomEnhanceParam>(attrs.parsed);
+
+ Stream<cpu> *s = ctx.get_stream<cpu>();
+ Random<cpu> *prnd = ctx.requested[0].get_random<cpu, real_t>(s);
+ float alpha = std::uniform_real_distribution<float>(
+ param.min_factor, param.max_factor)(prnd->GetRndEngine());
+
+ AdjustHueImpl(alpha, ctx, inputs, req, outputs);
}
-static void RandomColorJitter(const nnvm::NodeAttrs &attrs,
- const OpContext &ctx,
- const std::vector<TBlob> &inputs,
- const std::vector<OpReqType> &req,
- const std::vector<TBlob> &outputs) {
+struct RandomColorJitterParam : public dmlc::Parameter<RandomColorJitterParam> {
+ float brightness;
+ float contrast;
+ float saturation;
+ float hue;
+ DMLC_DECLARE_PARAMETER(RandomColorJitterParam) {
+ DMLC_DECLARE_FIELD(brightness)
+ .describe("How much to jitter brightness.");
+ DMLC_DECLARE_FIELD(contrast)
+ .describe("How much to jitter contrast.");
+ DMLC_DECLARE_FIELD(saturation)
+ .describe("How much to jitter saturation.");
+ DMLC_DECLARE_FIELD(hue)
+ .describe("How much to jitter hue.");
+ }
+};
+
+void RandomColorJitter(const nnvm::NodeAttrs &attrs,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+ using namespace mshadow;
+ const RandomColorJitterParam ¶m = nnvm::get<RandomColorJitterParam>(attrs.parsed);
+ Stream<cpu> *s = ctx.get_stream<cpu>();
+ Random<cpu> *prnd = ctx.requested[0].get_random<cpu, real_t>(s);
+
+ int order[4] = {0, 1, 2, 3};
+ std::shuffle(order, order + 4, prnd->GetRndEngine());
+ bool flag = false;
+
+ for (int i = 0; i < 4; ++i) {
+ switch (order[i]) {
+ case 0:
+ if (param.brightness > 0) {
+ float alpha_b = 1.0 + std::uniform_real_distribution<float>(
+ -param.brightness, param.brightness)(prnd->GetRndEngine());
+ AdjustBrightnessImpl(alpha_b, ctx, flag ? outputs : inputs, req, outputs);
+ flag = true;
+ }
+ break;
+ case 1:
+ if (param.contrast > 0) {
+ float alpha_c = 1.0 + std::uniform_real_distribution<float>(
+ -param.contrast, param.contrast)(prnd->GetRndEngine());
+ AdjustContrastImpl(alpha_c, ctx, flag ? outputs : inputs, req, outputs);
+ flag = true;
+ }
+ break;
+ case 2:
+ if (param.saturation > 0) {
+ float alpha_s = 1.f + std::uniform_real_distribution<float>(
+ -param.saturation, param.saturation)(prnd->GetRndEngine());
+ AdjustSaturationImpl(alpha_s, ctx, flag ? outputs : inputs, req, outputs);
+ flag = true;
+ }
+ break;
+ case 3:
+ if (param.hue > 0) {
+ float alpha_h = std::uniform_real_distribution<float>(
+ -param.hue, param.hue)(prnd->GetRndEngine());
+ AdjustHueImpl(alpha_h, ctx, flag ? outputs : inputs, req, outputs);
+ flag = true;
+ }
+ break;
+ }
+ }
}
struct AdjustLightingParam : public dmlc::Parameter<AdjustLightingParam> {
- nnvm::Tuple<float> alpha_rgb;
- nnvm::Tuple<float> eigval;
- nnvm::Tuple<float> eigvec;
+ nnvm::Tuple<float> alpha;
DMLC_DECLARE_PARAMETER(AdjustLightingParam) {
- DMLC_DECLARE_FIELD(alpha_rgb)
- .set_default({0, 0, 0})
+ DMLC_DECLARE_FIELD(alpha)
.describe("The lighting alphas for the R, G, B channels.");
- DMLC_DECLARE_FIELD(eigval)
- .describe("Eigen value.")
- .set_default({ 55.46, 4.794, 1.148 });
- DMLC_DECLARE_FIELD(eigvec)
- .describe("Eigen vector.")
- .set_default({ -0.5675, 0.7192, 0.4009,
- -0.5808, -0.0045, -0.8140,
- -0.5808, -0.0045, -0.8140 });
}
};
struct RandomLightingParam : public dmlc::Parameter<RandomLightingParam> {
float alpha_std;
- nnvm::Tuple<float> eigval;
- nnvm::Tuple<float> eigvec;
DMLC_DECLARE_PARAMETER(RandomLightingParam) {
DMLC_DECLARE_FIELD(alpha_std)
.set_default(0.05)
.describe("Level of the lighting noise.");
- DMLC_DECLARE_FIELD(eigval)
- .describe("Eigen value.")
- .set_default({ 55.46, 4.794, 1.148 });
- DMLC_DECLARE_FIELD(eigvec)
- .describe("Eigen vector.")
- .set_default({ -0.5675, 0.7192, 0.4009,
- -0.5808, -0.0045, -0.8140,
- -0.5808, -0.0045, -0.8140 });
}
};
-void AdjustLightingImpl(uint8_t* dst, const uint8_t* src,
- float alpha_r, float alpha_g, float alpha_b,
- const nnvm::Tuple<float> eigval, const nnvm::Tuple<float> eigvec,
- int H, int W) {
- alpha_r *= eigval[0];
- alpha_g *= eigval[1];
- alpha_b *= eigval[2];
- float pca_r = alpha_r * eigvec[0] + alpha_g * eigvec[1] + alpha_b * eigvec[2];
- float pca_g = alpha_r * eigvec[3] + alpha_g * eigvec[4] + alpha_b * eigvec[5];
- float pca_b = alpha_r * eigvec[6] + alpha_g * eigvec[7] + alpha_b * eigvec[8];
- for (int i = 0; i < H * W; i++) {
- int base_ind = 3 * i;
- float in_r = static_cast<float>(src[base_ind]);
- float in_g = static_cast<float>(src[base_ind + 1]);
- float in_b = static_cast<float>(src[base_ind + 2]);
- dst[base_ind] = std::min(255, std::max(0, static_cast<int>(in_r + pca_r)));
- dst[base_ind + 1] = std::min(255, std::max(0, static_cast<int>(in_g + pca_g)));
- dst[base_ind + 2] = std::min(255, std::max(0, static_cast<int>(in_b + pca_b)));
+void AdjustLightingImpl(const nnvm::Tuple<float>& alpha,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+ static const float eig[3][3] = {
+ { 55.46 * -0.5675, 4.794 * 0.7192, 1.148 * 0.4009 },
+ { 55.46 * -0.5808, 4.794 * -0.0045, 1.148 * -0.8140 },
+ { 55.46 * -0.5836, 4.794 * -0.6948, 1.148 * 0.4203 }
+ };
+
+ int length = inputs[0].shape_[0] * inputs[0].shape_[1];
+ int channels = inputs[0].shape_[2];
+ if (channels == 1) return;
+
+ float pca_r = eig[0][0] * alpha[0] + eig[0][1] * alpha[1] + eig[0][2] * alpha[2];
+ float pca_g = eig[1][0] * alpha[0] + eig[1][1] * alpha[1] + eig[1][2] * alpha[2];
+ float pca_b = eig[2][0] * alpha[0] + eig[2][1] * alpha[1] + eig[2][2] * alpha[2];
+
+ MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
+ DType* output = outputs[0].dptr<DType>();
+ DType* input = inputs[0].dptr<DType>();
+
+ for (int i = 0; i < length; i++) {
+ int base_ind = 3 * i;
+ float in_r = static_cast<float>(input[base_ind]);
+ float in_g = static_cast<float>(input[base_ind + 1]);
+ float in_b = static_cast<float>(input[base_ind + 2]);
+ output[base_ind] = saturate_cast<DType>(in_r + pca_r);
+ output[base_ind + 1] = saturate_cast<DType>(in_g + pca_g);
+ output[base_ind + 2] = saturate_cast<DType>(in_b + pca_b);
}
+ });
}
-static void AdjustLighting(const nnvm::NodeAttrs &attrs,
- const OpContext &ctx,
- const std::vector<TBlob> &inputs,
- const std::vector<OpReqType> &req,
- const std::vector<TBlob> &outputs) {
- using namespace mshadow;
- const AdjustLightingParam ¶m = nnvm::get<AdjustLightingParam>(attrs.parsed);
- CHECK_EQ(param.eigval.ndim(), 3) << "There should be 3 numbers in the eigval.";
- CHECK_EQ(param.eigvec.ndim(), 9) << "There should be 9 numbers in the eigvec.";
- CHECK_EQ(inputs[0].ndim(), 3);
- CHECK_EQ(inputs[0].size(2), 3);
- int H = inputs[0].size(0);
- int W = inputs[0].size(1);
- AdjustLightingImpl(outputs[0].dptr<uint8_t>(), inputs[0].dptr<uint8_t>(),
- param.alpha_rgb[0], param.alpha_rgb[1], param.alpha_rgb[2],
- param.eigval, param.eigvec, H, W);
+void AdjustLighting(const nnvm::NodeAttrs &attrs,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+ using namespace mshadow;
+ const AdjustLightingParam ¶m = nnvm::get<AdjustLightingParam>(attrs.parsed);
+ AdjustLightingImpl(param.alpha, ctx, inputs, req, outputs);
}
-static void RandomLighting(const nnvm::NodeAttrs &attrs,
- const OpContext &ctx,
- const std::vector<TBlob> &inputs,
- const std::vector<OpReqType> &req,
- const std::vector<TBlob> &outputs) {
- using namespace mshadow;
- const RandomLightingParam ¶m = nnvm::get<RandomLightingParam>(attrs.parsed);
- CHECK_EQ(param.eigval.ndim(), 3) << "There should be 3 numbers in the eigval.";
- CHECK_EQ(param.eigvec.ndim(), 9) << "There should be 9 numbers in the eigvec.";
- CHECK_EQ(inputs[0].ndim(), 3);
- CHECK_EQ(inputs[0].size(2), 3);
- int H = inputs[0].size(0);
- int W = inputs[0].size(1);
- Stream<cpu> *s = ctx.get_stream<cpu>();
- Random<cpu> *prnd = ctx.requested[0].get_random<cpu, real_t>(s);
- std::normal_distribution<float> dist(0, param.alpha_std);
- float alpha_r = dist(prnd->GetRndEngine());
- float alpha_g = dist(prnd->GetRndEngine());
- float alpha_b = dist(prnd->GetRndEngine());
- AdjustLightingImpl(outputs[0].dptr<uint8_t>(), inputs[0].dptr<uint8_t>(),
- alpha_r, alpha_g, alpha_b,
- param.eigval, param.eigvec, H, W);
+void RandomLighting(const nnvm::NodeAttrs &attrs,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+ using namespace mshadow;
+ const RandomLightingParam ¶m = nnvm::get<RandomLightingParam>(attrs.parsed);
+ Stream<cpu> *s = ctx.get_stream<cpu>();
+ Random<cpu> *prnd = ctx.requested[0].get_random<cpu, float>(s);
+ std::normal_distribution<float> dist(0, param.alpha_std);
+ float alpha_r = dist(prnd->GetRndEngine());
+ float alpha_g = dist(prnd->GetRndEngine());
+ float alpha_b = dist(prnd->GetRndEngine());
+ AdjustLightingImpl({alpha_r, alpha_g, alpha_b}, ctx, inputs, req, outputs);
}
-
-
-
+} // namespace image
} // namespace op
} // namespace mxnet
diff --git a/src/operator/image/image_random.cc b/src/operator/image/image_random.cc
index 29edeed..5a21bf8 100644
--- a/src/operator/image/image_random.cc
+++ b/src/operator/image/image_random.cc
@@ -30,6 +30,13 @@
namespace mxnet {
namespace op {
+namespace image {
+
+DMLC_REGISTER_PARAMETER(NormalizeParam);
+DMLC_REGISTER_PARAMETER(RandomEnhanceParam);
+DMLC_REGISTER_PARAMETER(AdjustLightingParam);
+DMLC_REGISTER_PARAMETER(RandomLightingParam);
+DMLC_REGISTER_PARAMETER(RandomColorJitterParam);
NNVM_REGISTER_OP(_image_to_tensor)
.describe(R"code()code" ADD_FILELINE)
@@ -42,13 +49,12 @@ NNVM_REGISTER_OP(_image_to_tensor)
.add_argument("data", "NDArray-or-Symbol", "The input.");
-DMLC_REGISTER_PARAMETER(NormalizeParam);
NNVM_REGISTER_OP(_image_normalize)
.describe(R"code()code" ADD_FILELINE)
.set_num_inputs(1)
.set_num_outputs(1)
.set_attr_parser(ParamParser<NormalizeParam>)
-.set_attr<nnvm::FInferShape>("FInferShape", ElemwiseShape<1, 1>)
+.set_attr<nnvm::FInferShape>("FInferShape", NormalizeShape)
.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
.set_attr<nnvm::FInplaceOption>("FInplaceOption",
[](const NodeAttrs& attrs){
@@ -59,33 +65,31 @@ NNVM_REGISTER_OP(_image_normalize)
.add_argument("data", "NDArray-or-Symbol", "The input.")
.add_arguments(NormalizeParam::__FIELDS__());
-DMLC_REGISTER_PARAMETER(FlipParam);
-NNVM_REGISTER_OP(_image_flip)
+
+NNVM_REGISTER_OP(_image_random_horizontal_flip)
.describe(R"code()code" ADD_FILELINE)
.set_num_inputs(1)
.set_num_outputs(1)
-.set_attr_parser(ParamParser<FlipParam>)
-.set_attr<nnvm::FInferShape>("FInferShape", ElemwiseShape<1, 1>)
+.set_attr<nnvm::FInferShape>("FInferShape", ImageShape)
.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
.set_attr<nnvm::FInplaceOption>("FInplaceOption",
- [](const NodeAttrs& attrs){
- return std::vector<std::pair<int, int> >{{0, 0}};
- })
-.set_attr<FCompute>("FCompute<cpu>", Flip)
+ [](const NodeAttrs& attrs){
+ return std::vector<std::pair<int, int> >{{0, 0}};
+ })
+.set_attr<FCompute>("FCompute<cpu>", RandomHorizontalFlip)
.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
-.add_argument("data", "NDArray-or-Symbol", "The input.")
-.add_arguments(FlipParam::__FIELDS__());
+.add_argument("data", "NDArray-or-Symbol", "The input.");
+
-DMLC_REGISTER_PARAMETER(RandomBrightnessParam);
NNVM_REGISTER_OP(_image_random_brightness)
.describe(R"code()code" ADD_FILELINE)
.set_num_inputs(1)
.set_num_outputs(1)
-.set_attr_parser(ParamParser<RandomBrightnessParam>)
+.set_attr_parser(ParamParser<RandomEnhanceParam>)
.set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& attrs) {
return std::vector<ResourceRequest>{ResourceRequest::kRandom};
})
-.set_attr<nnvm::FInferShape>("FInferShape", ElemwiseShape<1, 1>)
+.set_attr<nnvm::FInferShape>("FInferShape", ImageShape)
.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
.set_attr<nnvm::FInplaceOption>("FInplaceOption",
[](const NodeAttrs& attrs){
@@ -94,18 +98,18 @@ NNVM_REGISTER_OP(_image_random_brightness)
.set_attr<FCompute>("FCompute<cpu>", RandomBrightness)
.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
.add_argument("data", "NDArray-or-Symbol", "The input.")
-.add_arguments(RandomBrightnessParam::__FIELDS__());
+.add_arguments(RandomEnhanceParam::__FIELDS__());
+
-DMLC_REGISTER_PARAMETER(RandomContrastParam);
NNVM_REGISTER_OP(_image_random_contrast)
.describe(R"code()code" ADD_FILELINE)
.set_num_inputs(1)
.set_num_outputs(1)
-.set_attr_parser(ParamParser<RandomContrastParam>)
+.set_attr_parser(ParamParser<RandomEnhanceParam>)
.set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& attrs) {
return std::vector<ResourceRequest>{ResourceRequest::kRandom};
})
-.set_attr<nnvm::FInferShape>("FInferShape", ElemwiseShape<1, 1>)
+.set_attr<nnvm::FInferShape>("FInferShape", ImageShape)
.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
.set_attr<nnvm::FInplaceOption>("FInplaceOption",
[](const NodeAttrs& attrs){
@@ -114,18 +118,18 @@ NNVM_REGISTER_OP(_image_random_contrast)
.set_attr<FCompute>("FCompute<cpu>", RandomContrast)
.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
.add_argument("data", "NDArray-or-Symbol", "The input.")
-.add_arguments(RandomContrastParam::__FIELDS__());
+.add_arguments(RandomEnhanceParam::__FIELDS__());
+
-DMLC_REGISTER_PARAMETER(RandomSaturationParam);
NNVM_REGISTER_OP(_image_random_saturation)
.describe(R"code()code" ADD_FILELINE)
.set_num_inputs(1)
.set_num_outputs(1)
-.set_attr_parser(ParamParser<RandomSaturationParam>)
+.set_attr_parser(ParamParser<RandomEnhanceParam>)
.set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& attrs) {
return std::vector<ResourceRequest>{ResourceRequest::kRandom};
})
-.set_attr<nnvm::FInferShape>("FInferShape", ElemwiseShape<1, 1>)
+.set_attr<nnvm::FInferShape>("FInferShape", ImageShape)
.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
.set_attr<nnvm::FInplaceOption>("FInplaceOption",
[](const NodeAttrs& attrs){
@@ -134,31 +138,44 @@ NNVM_REGISTER_OP(_image_random_saturation)
.set_attr<FCompute>("FCompute<cpu>", RandomSaturation)
.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
.add_argument("data", "NDArray-or-Symbol", "The input.")
-.add_arguments(RandomSaturationParam::__FIELDS__());
+.add_arguments(RandomEnhanceParam::__FIELDS__());
-DMLC_REGISTER_PARAMETER(RandomHueParam);
NNVM_REGISTER_OP(_image_random_hue)
.describe(R"code()code" ADD_FILELINE)
.set_num_inputs(1)
.set_num_outputs(1)
-.set_attr_parser(ParamParser<RandomHueParam>)
+.set_attr_parser(ParamParser<RandomEnhanceParam>)
.set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& attrs) {
return std::vector<ResourceRequest>{ResourceRequest::kRandom};
})
-.set_attr<nnvm::FInferShape>("FInferShape", ElemwiseShape<1, 1>)
+.set_attr<nnvm::FInferShape>("FInferShape", ImageShape)
.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
-.set_attr<FCompute>("FCompute<cpu>", RandomHue<cpu>)
+.set_attr<FCompute>("FCompute<cpu>", RandomHue)
.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
.add_argument("data", "NDArray-or-Symbol", "The input.")
-.add_arguments(RandomHueParam::__FIELDS__());
+.add_arguments(RandomEnhanceParam::__FIELDS__());
+
+NNVM_REGISTER_OP(_image_random_color_jitter)
+.describe(R"code()code" ADD_FILELINE)
+.set_num_inputs(1)
+.set_num_outputs(1)
+.set_attr_parser(ParamParser<RandomColorJitterParam>)
+.set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& attrs) {
+ return std::vector<ResourceRequest>{ResourceRequest::kRandom};
+})
+.set_attr<nnvm::FInferShape>("FInferShape", ImageShape)
+.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
+.set_attr<FCompute>("FCompute<cpu>", RandomColorJitter)
+.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
+.add_argument("data", "NDArray-or-Symbol", "The input.")
+.add_arguments(RandomColorJitterParam::__FIELDS__());
-DMLC_REGISTER_PARAMETER(AdjustLightingParam);
NNVM_REGISTER_OP(_image_adjust_lighting)
.describe(R"code(Adjust the lighting level of the input. Follow the AlexNet style.)code" ADD_FILELINE)
.set_num_inputs(1)
.set_num_outputs(1)
.set_attr_parser(ParamParser<AdjustLightingParam>)
-.set_attr<nnvm::FInferShape>("FInferShape", ElemwiseShape<1, 1>)
+.set_attr<nnvm::FInferShape>("FInferShape", ImageShape)
.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
.set_attr<nnvm::FInplaceOption>("FInplaceOption",
[](const NodeAttrs& attrs){
@@ -169,7 +186,7 @@ NNVM_REGISTER_OP(_image_adjust_lighting)
.add_argument("data", "NDArray-or-Symbol", "The input.")
.add_arguments(AdjustLightingParam::__FIELDS__());
-DMLC_REGISTER_PARAMETER(RandomLightingParam);
+
NNVM_REGISTER_OP(_image_random_lighting)
.describe(R"code(Randomly add PCA noise. Follow the AlexNet style.)code" ADD_FILELINE)
.set_num_inputs(1)
@@ -178,7 +195,7 @@ NNVM_REGISTER_OP(_image_random_lighting)
.set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& attrs) {
return std::vector<ResourceRequest>{ResourceRequest::kRandom};
})
-.set_attr<nnvm::FInferShape>("FInferShape", ElemwiseShape<1, 1>)
+.set_attr<nnvm::FInferShape>("FInferShape", ImageShape)
.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
.set_attr<nnvm::FInplaceOption>("FInplaceOption",
[](const NodeAttrs& attrs){
@@ -189,5 +206,6 @@ NNVM_REGISTER_OP(_image_random_lighting)
.add_argument("data", "NDArray-or-Symbol", "The input.")
.add_arguments(RandomLightingParam::__FIELDS__());
+} // namespace image
} // namespace op
} // namespace mxnet
--
To stop receiving notification emails like this one, please contact
"commits@mxnet.apache.org" <co...@mxnet.apache.org>.
[incubator-mxnet] 11/19: Revert "add test script"
Posted by jx...@apache.org.
This is an automated email from the ASF dual-hosted git repository.
jxie pushed a commit to branch vision
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git
commit 400c380881e98291c47055ceaf89933a3debaa72
Author: Xingjian Shi <xs...@ust.hk>
AuthorDate: Tue Nov 21 11:08:23 2017 -0800
Revert "add test script"
This reverts commit 23f68272e305103ad87d089e700ef715b13067c0.
---
test_new_image_loader.py | 34 ----------------------------------
1 file changed, 34 deletions(-)
diff --git a/test_new_image_loader.py b/test_new_image_loader.py
deleted file mode 100644
index 296869e..0000000
--- a/test_new_image_loader.py
+++ /dev/null
@@ -1,34 +0,0 @@
-import os
-os.environ['MXNET_CPU_WORKER_NTHREADS'] = '1'
-os.environ['OMP_NUM_THREADS'] = '1'
-import time
-import numpy as np
-import multiprocessing as mp
-import mxnet as mx
-from mxnet import gluon as gl
-from mxnet.gluon.data.vision import transforms
-
-if __name__ == '__main__':
- M = 24
- BS = 100
-
- dataset = gl.data.vision.ImageFolderDataset('../256_ObjectCategories')
- transform = transforms.Compose([transforms.ToTensor(),
- transforms.RandomBrightness(1.0),
- transforms.RandomContrast(1.0),
- transforms.RandomSaturation(1.0),
- transforms.Normalize([0, 0, 0], [1, 1, 1])])
- dataset = dataset.transform_first(lambda x: transform(mx.image.center_crop(x, (224, 224))[0]))
- data_loader = gl.data.DataLoader(dataset, BS, shuffle=True, num_workers=M)
-
- N = len(dataset)
-
- iterator = iter(data_loader)
-
- tic = time.time()
-
- for data, label in iterator:
- data.wait_to_read()
- print(data.shape)
-
- print(N/(time.time() - tic))
--
To stop receiving notification emails like this one, please contact
"commits@mxnet.apache.org" <co...@mxnet.apache.org>.
[incubator-mxnet] 14/19: image flip op (#8759)
Posted by jx...@apache.org.
This is an automated email from the ASF dual-hosted git repository.
jxie pushed a commit to branch vision
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git
commit e97ef363bbae809f39c07415a6c6740b884c76f6
Author: Yizhi Liu <ja...@gmail.com>
AuthorDate: Sun Nov 26 17:20:55 2017 -0800
image flip op (#8759)
* image flip op
* rm image_common.h
* fix
* lint code
* flip optimize
---
src/operator/image/image_random-inl.h | 66 +++++++++++++++++++++++++++++++++--
src/operator/image/image_random.cc | 16 +++++++++
2 files changed, 79 insertions(+), 3 deletions(-)
diff --git a/src/operator/image/image_random-inl.h b/src/operator/image/image_random-inl.h
index ebbf60a..5c552b2 100644
--- a/src/operator/image/image_random-inl.h
+++ b/src/operator/image/image_random-inl.h
@@ -28,14 +28,20 @@
#include <mxnet/base.h>
#include <algorithm>
#include <vector>
-#include <opencv2/opencv.hpp>
-#include <opencv2/core/mat.hpp>
+#include <algorithm>
+#include <utility>
#include "../mxnet_op.h"
#include "../operator_common.h"
namespace mxnet {
namespace op {
+inline bool CheckIsImage(const TBlob &image) {
+ CHECK_EQ(image.type_flag_, mshadow::kUint8) << "input type is not an image.";
+ CHECK_EQ(image.ndim(), 3) << "input dimension is not 3.";
+ CHECK(image.shape_[2] == 1 || image.shape_[2] == 3) << "image channel should be 1 or 3.";
+}
+
static void RandomFlip(const nnvm::NodeAttrs &attrs,
const OpContext &ctx,
const std::vector<TBlob> &inputs,
@@ -76,6 +82,7 @@ static void ToTensor(const nnvm::NodeAttrs &attrs,
const std::vector<TBlob> &outputs) {
CHECK_EQ(req[0], kWriteTo)
<< "`to_tensor` does not support inplace";
+ CheckIsImage(inputs[0]);
int length = inputs[0].shape_[0] * inputs[0].shape_[1];
int channel = inputs[0].shape_[2];
@@ -101,7 +108,6 @@ struct NormalizeParam : public dmlc::Parameter<NormalizeParam> {
}
};
-
inline bool NormalizeShape(const nnvm::NodeAttrs& attrs,
std::vector<TShape> *in_attrs,
std::vector<TShape> *out_attrs) {
@@ -145,6 +151,60 @@ static void Normalize(const nnvm::NodeAttrs &attrs,
});
}
+struct FlipParam : public dmlc::Parameter<FlipParam> {
+ int axis;
+ DMLC_DECLARE_PARAMETER(FlipParam) {
+ DMLC_DECLARE_FIELD(axis)
+ .describe("0 or 1. 0 for horizontal flip, 1 for vertical flip.");
+ }
+};
+
+#define SWAP_IF_INPLACE(dst, dst_idx, src, src_idx) \
+ if (dst == src) { \
+ std::swap(dst[dst_idx], src[src_idx]); \
+ } else { \
+ dst[dst_idx] = src[src_idx]; \
+ }
+
+template<typename DType>
+static void FlipImpl(const TShape &shape, DType *src, DType *dst, int axis) {
+ const int height = shape[0];
+ const int width = shape[1];
+ const int nchannel = shape[2];
+
+ const int length = width * nchannel;
+ const int height_stride = (src == dst && axis == 1) ? (height >> 1) : height;
+ const int width_stride = (src == dst && axis == 0) ? (width >> 1) : width;
+
+ for (int h = 0; h < height_stride; ++h) {
+ const int h_dst = (axis == 0) ? h : (height - h);
+ for (int w = 0; w < width_stride; ++w) {
+ const int w_dst = (axis == 0) ? (width - w) : w;
+ const int idx_dst = h_dst * length + w_dst * nchannel;
+ const int idx_src = h * length + w * nchannel;
+ SWAP_IF_INPLACE(dst, idx_dst, src, idx_src);
+ if (nchannel > 1) {
+ SWAP_IF_INPLACE(dst, idx_dst + 1, src, idx_src + 1);
+ SWAP_IF_INPLACE(dst, idx_dst + 2, src, idx_src + 2);
+ }
+ }
+ }
+}
+
+static void Flip(const nnvm::NodeAttrs &attrs,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+ const FlipParam ¶m = nnvm::get<FlipParam>(attrs.parsed);
+ CHECK(param.axis == 0 || param.axis == 1) << "flip axis must be 0 or 1.";
+ CheckIsImage(inputs[0]);
+ const TShape& ishape = inputs[0].shape_;
+ MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
+ FlipImpl(ishape, inputs[0].dptr<DType>(), outputs[0].dptr<DType>(), param.axis);
+ });
+}
+
struct RandomBrightnessParam : public dmlc::Parameter<RandomBrightnessParam> {
float max_brightness;
DMLC_DECLARE_PARAMETER(RandomBrightnessParam) {
diff --git a/src/operator/image/image_random.cc b/src/operator/image/image_random.cc
index 5b47f50..4184382 100644
--- a/src/operator/image/image_random.cc
+++ b/src/operator/image/image_random.cc
@@ -59,6 +59,22 @@ NNVM_REGISTER_OP(_image_normalize)
.add_argument("data", "NDArray-or-Symbol", "The input.")
.add_arguments(NormalizeParam::__FIELDS__());
+DMLC_REGISTER_PARAMETER(FlipParam);
+NNVM_REGISTER_OP(_image_flip)
+.describe(R"code()code" ADD_FILELINE)
+.set_num_inputs(1)
+.set_num_outputs(1)
+.set_attr_parser(ParamParser<FlipParam>)
+.set_attr<nnvm::FInferShape>("FInferShape", ElemwiseShape<1, 1>)
+.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
+.set_attr<nnvm::FInplaceOption>("FInplaceOption",
+ [](const NodeAttrs& attrs){
+ return std::vector<std::pair<int, int> >{{0, 0}};
+ })
+.set_attr<FCompute>("FCompute<cpu>", Flip)
+.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
+.add_argument("data", "NDArray-or-Symbol", "The input.")
+.add_arguments(FlipParam::__FIELDS__());
DMLC_REGISTER_PARAMETER(RandomBrightnessParam);
NNVM_REGISTER_OP(_image_random_brightness)
--
To stop receiving notification emails like this one, please contact
"commits@mxnet.apache.org" <co...@mxnet.apache.org>.
[incubator-mxnet] 01/19: [WIP]Image Augmenter (#8633)
Posted by jx...@apache.org.
This is an automated email from the ASF dual-hosted git repository.
jxie pushed a commit to branch vision
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git
commit fa7575a53ef0f05d80e6e57f72d51123b06d0662
Author: Hu Shiwen <ya...@gmail.com>
AuthorDate: Tue Nov 14 08:32:47 2017 +0800
[WIP]Image Augmenter (#8633)
* add file
* add random_brightness
add python mx.sym/nd.image
* fix lint
* add image/image_common.h
* add RandomContrast
* change name
* fix
---
python/mxnet/base.py | 8 +-
python/mxnet/ndarray/__init__.py | 4 +-
.../mxnet/{symbol/__init__.py => ndarray/image.py} | 15 ++--
python/mxnet/symbol/__init__.py | 4 +-
python/mxnet/symbol/{__init__.py => image.py} | 15 ++--
src/operator/batch_norm_v1-inl.h | 2 +-
src/operator/image/image_common.h | 88 +++++++++++++++++++
src/operator/image/image_random-inl.h | 99 ++++++++++++++++++++++
src/operator/image/image_random.cc | 50 +++++++++++
src/operator/random/multisample_op.h | 2 +-
src/operator/tensor/broadcast_reduce_op_index.cc | 2 +-
.../elemwise_binary_broadcast_op_extended.cc | 2 +-
.../tensor/elemwise_binary_broadcast_op_logic.cc | 2 +-
13 files changed, 261 insertions(+), 32 deletions(-)
diff --git a/python/mxnet/base.py b/python/mxnet/base.py
index fbdf15f..7bdfb5b 100644
--- a/python/mxnet/base.py
+++ b/python/mxnet/base.py
@@ -449,7 +449,7 @@ def _as_list(obj):
return [obj]
-_OP_NAME_PREFIX_LIST = ['_contrib_', '_linalg_', '_sparse_']
+_OP_NAME_PREFIX_LIST = ['_contrib_', '_linalg_', '_sparse_', '_image_']
def _get_op_name_prefix(op_name):
@@ -503,10 +503,11 @@ def _init_op_module(root_namespace, module_name, make_op_func):
hdl = OpHandle()
check_call(_LIB.NNGetOpHandle(c_str(name), ctypes.byref(hdl)))
op_name_prefix = _get_op_name_prefix(name)
+ module_name_local = module_name
if len(op_name_prefix) > 0:
func_name = name[len(op_name_prefix):]
cur_module = submodule_dict[op_name_prefix]
- module_name = "%s.%s.%s" % (root_namespace, module_name, op_name_prefix[1:-1])
+ module_name_local = "%s.%s.%s" % (root_namespace, module_name, op_name_prefix[1:-1])
elif name.startswith('_'):
func_name = name
cur_module = module_internal
@@ -515,10 +516,11 @@ def _init_op_module(root_namespace, module_name, make_op_func):
cur_module = module_op
function = make_op_func(hdl, name, func_name)
- function.__module__ = module_name
+ function.__module__ = module_name_local
setattr(cur_module, function.__name__, function)
cur_module.__all__.append(function.__name__)
+
if op_name_prefix == '_contrib_':
hdl = OpHandle()
check_call(_LIB.NNGetOpHandle(c_str(name), ctypes.byref(hdl)))
diff --git a/python/mxnet/ndarray/__init__.py b/python/mxnet/ndarray/__init__.py
index 586dc9e..86a3a20 100644
--- a/python/mxnet/ndarray/__init__.py
+++ b/python/mxnet/ndarray/__init__.py
@@ -17,7 +17,7 @@
"""NDArray API of MXNet."""
-from . import _internal, contrib, linalg, op, random, sparse, utils
+from . import _internal, contrib, linalg, op, random, sparse, utils, image
# pylint: disable=wildcard-import, redefined-builtin
try:
from .gen_op import * # pylint: disable=unused-wildcard-import
@@ -31,4 +31,4 @@ from .utils import load, save, zeros, empty, array
from .sparse import _ndarray_cls
from .ndarray import _GRAD_REQ_MAP
-__all__ = op.__all__ + ndarray.__all__ + utils.__all__ + ['contrib', 'linalg', 'random', 'sparse']
+__all__ = op.__all__ + ndarray.__all__ + utils.__all__ + ['contrib', 'linalg', 'random', 'sparse', 'image']
diff --git a/python/mxnet/symbol/__init__.py b/python/mxnet/ndarray/image.py
similarity index 67%
copy from python/mxnet/symbol/__init__.py
copy to python/mxnet/ndarray/image.py
index a07025e..0afab24 100644
--- a/python/mxnet/symbol/__init__.py
+++ b/python/mxnet/ndarray/image.py
@@ -15,17 +15,12 @@
# specific language governing permissions and limitations
# under the License.
-"""Symbol API of MXNet."""
-
-from . import _internal, contrib, linalg, op, random, sparse
-# pylint: disable=wildcard-import, redefined-builtin
+# coding: utf-8
+# pylint: disable=wildcard-import, unused-wildcard-import
+"""Image NDArray API of MXNet."""
try:
- from .gen_op import * # pylint: disable=unused-wildcard-import
+ from .gen_iamge import *
except ImportError:
pass
-from . import register
-from .op import *
-from .symbol import *
-# pylint: enable=wildcard-import
-__all__ = op.__all__ + symbol.__all__ + ['contrib', 'linalg', 'random', 'sparse']
+__all__ = []
diff --git a/python/mxnet/symbol/__init__.py b/python/mxnet/symbol/__init__.py
index a07025e..a10b64e 100644
--- a/python/mxnet/symbol/__init__.py
+++ b/python/mxnet/symbol/__init__.py
@@ -17,7 +17,7 @@
"""Symbol API of MXNet."""
-from . import _internal, contrib, linalg, op, random, sparse
+from . import _internal, contrib, linalg, op, random, sparse, image
# pylint: disable=wildcard-import, redefined-builtin
try:
from .gen_op import * # pylint: disable=unused-wildcard-import
@@ -28,4 +28,4 @@ from .op import *
from .symbol import *
# pylint: enable=wildcard-import
-__all__ = op.__all__ + symbol.__all__ + ['contrib', 'linalg', 'random', 'sparse']
+__all__ = op.__all__ + symbol.__all__ + ['contrib', 'linalg', 'random', 'sparse', 'image']
diff --git a/python/mxnet/symbol/__init__.py b/python/mxnet/symbol/image.py
similarity index 67%
copy from python/mxnet/symbol/__init__.py
copy to python/mxnet/symbol/image.py
index a07025e..7624bcc 100644
--- a/python/mxnet/symbol/__init__.py
+++ b/python/mxnet/symbol/image.py
@@ -15,17 +15,12 @@
# specific language governing permissions and limitations
# under the License.
-"""Symbol API of MXNet."""
-
-from . import _internal, contrib, linalg, op, random, sparse
-# pylint: disable=wildcard-import, redefined-builtin
+# coding: utf-8
+# pylint: disable=wildcard-import, unused-wildcard-import
+"""Image Symbol API of MXNet."""
try:
- from .gen_op import * # pylint: disable=unused-wildcard-import
+ from .gen_iamge import *
except ImportError:
pass
-from . import register
-from .op import *
-from .symbol import *
-# pylint: enable=wildcard-import
-__all__ = op.__all__ + symbol.__all__ + ['contrib', 'linalg', 'random', 'sparse']
+__all__ = []
diff --git a/src/operator/batch_norm_v1-inl.h b/src/operator/batch_norm_v1-inl.h
index 329d66d..1e04845 100644
--- a/src/operator/batch_norm_v1-inl.h
+++ b/src/operator/batch_norm_v1-inl.h
@@ -19,7 +19,7 @@
/*!
* Copyright (c) 2015 by Contributors
- * \file batch_norm-inl_v1.h
+ * \file batch_norm_v1-inl.h
* \brief
* \author Bing Xu
*/
diff --git a/src/operator/image/image_common.h b/src/operator/image/image_common.h
new file mode 100644
index 0000000..7cf3f96
--- /dev/null
+++ b/src/operator/image/image_common.h
@@ -0,0 +1,88 @@
+/*
+* Licensed to the Apache Software Foundation (ASF) under one
+* or more contributor license agreements. See the NOTICE file
+* distributed with this work for additional information
+* regarding copyright ownership. The ASF licenses this file
+* to you under the Apache License, Version 2.0 (the
+* "License"); you may not use this file except in compliance
+* with the License. You may obtain a copy of the License at
+*
+* http://www.apache.org/licenses/LICENSE-2.0
+*
+* Unless required by applicable law or agreed to in writing,
+* software distributed under the License is distributed on an
+* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+* KIND, either express or implied. See the License for the
+* specific language governing permissions and limitations
+* under the License.
+*/
+
+/*!
+* \file image_common.h
+* \brief
+* \author
+*/
+#ifndef MXNET_OPERATOR_IMAGE_IMAGE_COMMON_H_
+#define MXNET_OPERATOR_IMAGE_IMAGE_COMMON_H_
+
+#include <mxnet/base.h>
+
+namespace mxnet {
+namespace op {
+
+/**
+* @brief convert TBlob to cv::Mat
+* @param input @see TBlob
+* @param hight
+* @param weight
+* @param channel
+* @return
+*/
+static cv::Mat mat_convert(TBlob input, int hight, int weight, int channel) {
+ cv::Mat m;
+ switch (input.type_flag_) {
+ case mshadow::kFloat32: {
+ typedef float DType;
+ m = cv::Mat(hight, weight, CV_MAKETYPE(CV_32F, channel), input.dptr<DType>());
+ }
+ break;
+ case mshadow::kFloat64: {
+ typedef double DType;
+ m = cv::Mat(hight, weight, CV_MAKETYPE(CV_64F, channel), input.dptr<DType>());
+ }
+ break;
+ case mshadow::kFloat16: {
+ typedef mshadow::half::half_t DType;
+ LOG(FATAL) << "not support type enum " << input.type_flag_;
+ }
+ break;
+ case mshadow::kUint8: {
+ typedef uint8_t DType;
+ m = cv::Mat(hight, weight, CV_MAKETYPE(CV_8U, channel), input.dptr<DType>());
+ }
+ break;
+ case mshadow::kInt8: {
+ typedef int8_t DType;
+ m = cv::Mat(hight, weight, CV_MAKETYPE(CV_8S, channel), input.dptr<DType>());
+ }
+ break;
+ case mshadow::kInt32: {
+ typedef int32_t DType;
+ m = cv::Mat(hight, weight, CV_MAKETYPE(CV_32S, channel), input.dptr<DType>());
+ }
+ break;
+ case mshadow::kInt64: {
+ typedef int64_t DType;
+ LOG(FATAL) << "not support type enum " << input.type_flag_;
+ }
+ break;
+ default:
+ LOG(FATAL) << "Unknown type enum " << input.type_flag_;
+ }
+ return m;
+}
+} // namespace op
+} // namespace mxnet
+
+
+#endif // MXNET_OPERATOR_IMAGE_IMAGE_COMMON_H_
\ No newline at end of file
diff --git a/src/operator/image/image_random-inl.h b/src/operator/image/image_random-inl.h
new file mode 100644
index 0000000..027d587
--- /dev/null
+++ b/src/operator/image/image_random-inl.h
@@ -0,0 +1,99 @@
+/*
+* Licensed to the Apache Software Foundation (ASF) under one
+* or more contributor license agreements. See the NOTICE file
+* distributed with this work for additional information
+* regarding copyright ownership. The ASF licenses this file
+* to you under the Apache License, Version 2.0 (the
+* "License"); you may not use this file except in compliance
+* with the License. You may obtain a copy of the License at
+*
+* http://www.apache.org/licenses/LICENSE-2.0
+*
+* Unless required by applicable law or agreed to in writing,
+* software distributed under the License is distributed on an
+* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+* KIND, either express or implied. See the License for the
+* specific language governing permissions and limitations
+* under the License.
+*/
+
+/*!
+* \file image_random-inl.h
+* \brief
+* \author
+*/
+#ifndef MXNET_OPERATOR_IMAGE_IMAGE_RANDOM_INL_H_
+#define MXNET_OPERATOR_IMAGE_IMAGE_RANDOM_INL_H_
+
+#include <vector>
+#include <mxnet/base.h>
+#include <opencv2/opencv.hpp>
+#include <opencv2/core/mat.hpp>
+#include "mxnet/op_attr_types.h"
+#include "image_common.h"
+
+
+namespace mxnet {
+namespace op {
+struct RandomBrightnessParam : public dmlc::Parameter<RandomBrightnessParam> {
+ float max_brightness;
+ DMLC_DECLARE_PARAMETER(RandomBrightnessParam) {
+ DMLC_DECLARE_FIELD(max_brightness)
+ .set_default(0.0)
+ .describe("Max Contrast.");
+ }
+};
+
+
+template<typename xpu>
+static void RandomBrightness(const nnvm::NodeAttrs &attrs,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+ auto input = inputs[0];
+ auto output = outputs[0];
+ int hight = input.shape_[0];
+ int weight = input.shape_[1];
+ int channel = input.shape_[2];
+
+ auto input_mat = mat_convert(input, hight, weight, channel);
+ auto output_mat = mat_convert(output, hight, weight, channel);
+ //input_mat.convertTo(output_mat, -1, 1/255.0, 0);
+ std::default_random_engine generator;
+ const RandomBrightnessParam ¶m = nnvm::get<RandomBrightnessParam>(attrs.parsed);
+ float alpha_b = 1.0 + std::uniform_real_distribution<float>(-param.max_brightness, param.max_brightness)(generator);
+ output_mat.convertTo(output_mat, -1, alpha_b, 0);
+}
+
+
+template<typename xpu>
+static void RandomContrast(const nnvm::NodeAttrs &attrs,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+ auto input = inputs[0];
+ auto output = outputs[0];
+ int hight = input.shape_[0];
+ int weight = input.shape_[1];
+ int channel = input.shape_[2];
+
+ auto input_mat = mat_convert(input, hight, weight, channel);
+ auto output_mat = mat_convert(output, hight, weight, channel);
+ //input_mat.convertTo(output_mat, -1, 1/255.0, 0);
+ std::default_random_engine generator;
+ const RandomBrightnessParam ¶m = nnvm::get<RandomBrightnessParam>(attrs.parsed);
+ float alpha_c = 1.0 + std::uniform_real_distribution<float>(-param.max_brightness, param.max_brightness)(generator);
+ cv::Mat temp_;
+ cv::cvtColor(input_mat, temp_, CV_RGB2GRAY);
+ float gray_mean = cv::mean(temp_)[0];
+ input_mat.convertTo(output_mat, -1, alpha_c, (1 - alpha_c) * gray_mean);
+
+}
+
+
+} // namespace op
+} // namespace mxnet
+
+#endif // MXNET_OPERATOR_IMAGE_IMAGE_RANDOM_INL_H_
diff --git a/src/operator/image/image_random.cc b/src/operator/image/image_random.cc
new file mode 100644
index 0000000..3777e43
--- /dev/null
+++ b/src/operator/image/image_random.cc
@@ -0,0 +1,50 @@
+/*
+* Licensed to the Apache Software Foundation (ASF) under one
+* or more contributor license agreements. See the NOTICE file
+* distributed with this work for additional information
+* regarding copyright ownership. The ASF licenses this file
+* to you under the Apache License, Version 2.0 (the
+* "License"); you may not use this file except in compliance
+* with the License. You may obtain a copy of the License at
+*
+* http://www.apache.org/licenses/LICENSE-2.0
+*
+* Unless required by applicable law or agreed to in writing,
+* software distributed under the License is distributed on an
+* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+* KIND, either express or implied. See the License for the
+* specific language governing permissions and limitations
+* under the License.
+*/
+
+/*!
+* \file image_random.cc
+* \brief
+* \author
+*/
+
+#include <mxnet/base.h>
+#include "./image_random-inl.h"
+#include "operator/operator_common.h"
+#include "operator/elemwise_op_common.h"
+
+
+
+namespace mxnet {
+namespace op {
+
+DMLC_REGISTER_PARAMETER(RandomBrightnessParam);
+NNVM_REGISTER_OP(_image_random_brightness)
+.describe(R"code()code" ADD_FILELINE)
+.set_num_inputs(1)
+.set_num_outputs(1)
+.set_attr_parser(ParamParser<RandomBrightnessParam>)
+.set_attr<nnvm::FInferShape>("FInferShape", ElemwiseShape<1, 1>)
+.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
+.set_attr<FCompute>("FCompute<cpu>", RandomBrightness<cpu>)
+.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
+.add_argument("data", "NDArray-or-Symbol", "The input.")
+.add_arguments(RandomBrightnessParam::__FIELDS__());
+
+}
+}
diff --git a/src/operator/random/multisample_op.h b/src/operator/random/multisample_op.h
index e93e453..360c100 100644
--- a/src/operator/random/multisample_op.h
+++ b/src/operator/random/multisample_op.h
@@ -19,7 +19,7 @@
/*!
* Copyright (c) 2017 by Contributors
- * \file sampling_op.h
+ * \file multisample_op.h
* \brief Function definitions of operators for sampling from multiple distributions
*/
#ifndef MXNET_OPERATOR_RANDOM_MULTISAMPLE_OP_H_
diff --git a/src/operator/tensor/broadcast_reduce_op_index.cc b/src/operator/tensor/broadcast_reduce_op_index.cc
index dc07e67..6fd90df 100644
--- a/src/operator/tensor/broadcast_reduce_op_index.cc
+++ b/src/operator/tensor/broadcast_reduce_op_index.cc
@@ -19,7 +19,7 @@
/*!
* Copyright (c) 2016 by Contributors
- * \file broadcast_reduce_op.cc
+ * \file broadcast_reduce_op_index.cc
* \brief CPU Implementation of broadcast and reduce functions.
*/
#include "./broadcast_reduce_op.h"
diff --git a/src/operator/tensor/elemwise_binary_broadcast_op_extended.cc b/src/operator/tensor/elemwise_binary_broadcast_op_extended.cc
index fe7ad76..8fc3c48 100644
--- a/src/operator/tensor/elemwise_binary_broadcast_op_extended.cc
+++ b/src/operator/tensor/elemwise_binary_broadcast_op_extended.cc
@@ -19,7 +19,7 @@
/*!
* Copyright (c) 2016 by Contributors
- * \file elemwise_binary_scalar_op.cc
+ * \file elemwise_binary_broadcast_op_extended.cc
* \brief CPU Implementation of unary function.
*/
#include "./elemwise_unary_op.h"
diff --git a/src/operator/tensor/elemwise_binary_broadcast_op_logic.cc b/src/operator/tensor/elemwise_binary_broadcast_op_logic.cc
index 6d74f2d..31f34bb 100644
--- a/src/operator/tensor/elemwise_binary_broadcast_op_logic.cc
+++ b/src/operator/tensor/elemwise_binary_broadcast_op_logic.cc
@@ -19,7 +19,7 @@
/*!
* Copyright (c) 2016 by Contributors
- * \file elemwise_binary_scalar_op.cc
+ * \file elemwise_binary_broadcast_op_logic.cc
* \brief CPU Implementation of unary function.
*/
#include "./elemwise_unary_op.h"
--
To stop receiving notification emails like this one, please contact
"commits@mxnet.apache.org" <co...@mxnet.apache.org>.
[incubator-mxnet] 15/19: [WIP]hue (#8678)
Posted by jx...@apache.org.
This is an automated email from the ASF dual-hosted git repository.
jxie pushed a commit to branch vision
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git
commit 9e36e302aa8cb4b57ab467351e4db28ba63ba91f
Author: Hu Shiwen <ya...@gmail.com>
AuthorDate: Tue Nov 28 08:57:51 2017 +0800
[WIP]hue (#8678)
* add hue
* fix
* up sub
* fix lnit
---
src/operator/image/image_random-inl.h | 152 ++++++++++++++++++++++++++++++++++
src/operator/image/image_random.cc | 16 ++++
2 files changed, 168 insertions(+)
diff --git a/src/operator/image/image_random-inl.h b/src/operator/image/image_random-inl.h
index 5c552b2..3bee843 100644
--- a/src/operator/image/image_random-inl.h
+++ b/src/operator/image/image_random-inl.h
@@ -25,9 +25,12 @@
#ifndef MXNET_OPERATOR_IMAGE_IMAGE_RANDOM_INL_H_
#define MXNET_OPERATOR_IMAGE_IMAGE_RANDOM_INL_H_
+
#include <mxnet/base.h>
#include <algorithm>
#include <vector>
+#include <cmath>
+#include <limits>
#include <algorithm>
#include <utility>
#include "../mxnet_op.h"
@@ -337,11 +340,160 @@ static void RandomSaturation(const nnvm::NodeAttrs &attrs,
}
}
+struct RandomHueParam : public dmlc::Parameter<RandomHueParam> {
+ float max_hue;
+ DMLC_DECLARE_PARAMETER(RandomHueParam) {
+ DMLC_DECLARE_FIELD(max_hue)
+ .set_default(0.0)
+ .describe("Max Hue.");
+ }
+};
+
+template <typename DType> static
+void RGB2HLSConvert(const DType src_r,
+ const DType src_g,
+ const DType src_b,
+ DType *dst_h,
+ DType *dst_l,
+ DType *dst_s
+ ) {
+ DType b = src_b, g = src_g, r = src_r;
+ DType h = 0.f, s = 0.f, l;
+ DType vmin;
+ DType vmax;
+ DType diff;
+
+ vmax = vmin = r;
+ vmax = fmax(vmax, g);
+ vmax = fmax(vmax, b);
+ vmin = fmin(vmin, g);
+ vmin = fmin(vmin, b);
+
+ diff = vmax - vmin;
+ l = (vmax + vmin) * 0.5f;
+
+ if (diff > std::numeric_limits<DType>::epsilon()) {
+ s = (l < 0.5f) * diff / (vmax + vmin);
+ s += (l >= 0.5f) * diff / (2.0f - vmax - vmin);
+
+ diff = 60.f / diff;
+
+ h = (vmax == r) * (g - b) * diff;
+ h += (vmax != r && vmax == g) * ((b - r) * diff + 120.f);
+ h += (vmax != r && vmax != g) * ((r - g) * diff + 240.f);
+ h += (h < 0.f) * 360.f;
+ }
+
+ *dst_h = h;
+ *dst_l = l;
+ *dst_s = s;
+}
+
+
+static int c_HlsSectorData[6][3] = {
+ { 1, 3, 0 },
+ { 1, 0, 2 },
+ { 3, 0, 1 },
+ { 0, 2, 1 },
+ { 0, 1, 3 },
+ { 2, 1, 0 }
+};
+
+template <typename DType> static void HLS2RGBConvert(const DType src_h,
+ const DType src_l,
+ const DType src_s,
+ DType *dst_r,
+ DType *dst_g,
+ DType *dst_b) {
+
+
+ float h = src_h, l = src_l, s = src_s;
+ float b = l, g = l, r = l;
+
+ if (s != 0) {
+ float p2 = (l <= 0.5f) * l * (1 + s);
+ p2 += (l > 0.5f) * (l + s - l * s);
+ float p1 = 2 * l - p2;
+
+ if (h < 0) {
+ do { h += 6; } while (h < 0);
+ } else if (h >= 6) {
+ do { h -= 6; } while (h >= 6);
+ }
+
+ int sector = static_cast<int>(h);
+
+ h -= sector;
+
+ float tab[4];
+ tab[0] = p2;
+ tab[1] = p1;
+ tab[2] = p1 + (p2 - p1) * (1 - h);
+ tab[3] = p1 + (p2 - p1) * h;
+
+ b = tab[c_HlsSectorData[sector][0]];
+ g = tab[c_HlsSectorData[sector][1]];
+ r = tab[c_HlsSectorData[sector][2]];
+ }
+
+ *dst_b = b;
+ *dst_g = g;
+ *dst_r = r;
+}
+
+template<typename xpu, typename DType>
+static void RandomHueKernal(const TBlob &input,
+ const TBlob &output,
+ Stream<xpu> *s,
+ int hight,
+ int weight,
+ DType alpha) {
+ auto input_3d = input.get<xpu, 3, DType>(s);
+ auto output_3d = output.get<xpu, 3, DType>(s);
+ for (int h_index = 0; h_index < hight; ++h_index) {
+ for (int w_index = 0; w_index < weight; ++w_index) {
+ DType h;
+ DType l;
+ DType s;
+ RGB2HLSConvert(input_3d[0][h_index][w_index],
+ input_3d[1][h_index][w_index],
+ input_3d[2][h_index][w_index],
+ &h, &l, &s);
+ h += alpha;
+ h = std::max(DType(0), std::min(DType(180), h));
+
+ HLS2RGBConvert(
+ h, l, s,
+ &output_3d[0][h_index][w_index],
+ &output_3d[1][h_index][w_index],
+ &output_3d[2][h_index][w_index]);
+ }
+ }
+}
+
+template<typename xpu>
static void RandomHue(const nnvm::NodeAttrs &attrs,
const OpContext &ctx,
const std::vector<TBlob> &inputs,
const std::vector<OpReqType> &req,
const std::vector<TBlob> &outputs) {
+ using namespace mshadow;
+ auto input = inputs[0];
+ auto output = outputs[0];
+ int channel = input.shape_[0];
+ int hight = input.shape_[1];
+ int weight = input.shape_[2];
+ Stream<xpu> *s = ctx.get_stream<xpu>();
+ Random<xpu> *prnd = ctx.requested[kRandom].get_random<xpu, real_t>(s);
+
+ const RandomHueParam ¶m = nnvm::get<RandomHueParam>(attrs.parsed);
+ float alpha = std::uniform_real_distribution<float>(
+ -param.max_hue, param.max_hue)(prnd->GetRndEngine());
+ auto output_float = output.get<xpu, 3, float>(s);
+
+ MSHADOW_TYPE_SWITCH(input.type_flag_, DType, {
+ RandomHueKernal<xpu, DType>(input, output, s, hight, weight, alpha);
+ });
}
static void RandomColorJitter(const nnvm::NodeAttrs &attrs,
diff --git a/src/operator/image/image_random.cc b/src/operator/image/image_random.cc
index 4184382..29edeed 100644
--- a/src/operator/image/image_random.cc
+++ b/src/operator/image/image_random.cc
@@ -136,6 +136,22 @@ NNVM_REGISTER_OP(_image_random_saturation)
.add_argument("data", "NDArray-or-Symbol", "The input.")
.add_arguments(RandomSaturationParam::__FIELDS__());
+DMLC_REGISTER_PARAMETER(RandomHueParam);
+NNVM_REGISTER_OP(_image_random_hue)
+.describe(R"code()code" ADD_FILELINE)
+.set_num_inputs(1)
+.set_num_outputs(1)
+.set_attr_parser(ParamParser<RandomHueParam>)
+.set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& attrs) {
+ return std::vector<ResourceRequest>{ResourceRequest::kRandom};
+})
+.set_attr<nnvm::FInferShape>("FInferShape", ElemwiseShape<1, 1>)
+.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
+.set_attr<FCompute>("FCompute<cpu>", RandomHue<cpu>)
+.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
+.add_argument("data", "NDArray-or-Symbol", "The input.")
+.add_arguments(RandomHueParam::__FIELDS__());
+
DMLC_REGISTER_PARAMETER(AdjustLightingParam);
NNVM_REGISTER_OP(_image_adjust_lighting)
.describe(R"code(Adjust the lighting level of the input. Follow the AlexNet style.)code" ADD_FILELINE)
--
To stop receiving notification emails like this one, please contact
"commits@mxnet.apache.org" <co...@mxnet.apache.org>.
[incubator-mxnet] 08/19: [Image OP] Normalize (#8731)
Posted by jx...@apache.org.
This is an automated email from the ASF dual-hosted git repository.
jxie pushed a commit to branch vision
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git
commit a399908ace36caa1e2db3eff4e305b71bbf168d5
Author: Yizhi Liu <ja...@gmail.com>
AuthorDate: Mon Nov 20 11:57:22 2017 -0800
[Image OP] Normalize (#8731)
* image normalize op
* image normalize op param check
---
src/operator/image/image_random-inl.h | 93 +++++++++++++++++++++++++++++++++++
src/operator/image/image_random.cc | 34 ++++++++++++-
2 files changed, 125 insertions(+), 2 deletions(-)
diff --git a/src/operator/image/image_random-inl.h b/src/operator/image/image_random-inl.h
index c50ecb7..6f9cdc0 100644
--- a/src/operator/image/image_random-inl.h
+++ b/src/operator/image/image_random-inl.h
@@ -106,12 +106,105 @@ static void ToTensor(const nnvm::NodeAttrs &attrs,
});
}
+struct NormalizeParam : public dmlc::Parameter<NormalizeParam> {
+ nnvm::Tuple<float> mean;
+ nnvm::Tuple<float> std;
+ DMLC_DECLARE_PARAMETER(NormalizeParam) {
+ DMLC_DECLARE_FIELD(mean)
+ .describe("Sequence of mean for each channel.");
+ DMLC_DECLARE_FIELD(std)
+ .describe("Sequence of standard deviations for each channel.");
+ }
+};
+
+struct normalize {
+ template<typename DType>
+ MSHADOW_XINLINE static void Map(int i, DType *out, const DType *in,
+ const OpReqType req,
+ const int nchannel, const int size,
+ const float *mean, const float *std) {
+ int c = 0;
+ switch (nchannel) {
+ case 1:
+ break;
+ case 3:
+ if (i < size) {
+ c = 0;
+ } else if (i < (size << 1)) {
+ c = 1;
+ } else {
+ c = 2;
+ }
+ break;
+ default:
+ LOG(FATAL) << "not support channel" << nchannel;
+ }
+ float m = (mean ? mean[c] : 0);
+ KERNEL_ASSIGN(out[i], req, static_cast<DType>((in[i] - m) / std[c]));
+ }
+};
+
+static void NormalizeCheckParam(const nnvm::Tuple<float> &mean,
+ const nnvm::Tuple<float> &std,
+ const int nchannel) {
+ CHECK(mean.ndim() == 1 || mean.ndim() == 3)
+ << "Mean must be in dimension 1 or 3.";
+ CHECK(std.ndim() == 1 || std.ndim() == 3)
+ << "Standard deviations must be in dimension 1 or 3.";
+ CHECK(nchannel == 1 || nchannel == 3) << "Image channel must be 1 or 3.";
+ CHECK_EQ(mean.ndim(), nchannel)
+ << "Mean dimension does not agree with image channel.";
+ CHECK_EQ(std.ndim(), nchannel)
+ << "Standard deviations dimension does not agree with image channel.";
+ for (uint32_t c = 0; c < std.ndim(); ++c) {
+ CHECK(std[c] > 0) << "Invalid standard deviation " << std[c];
+ }
+}
+
template<typename xpu>
static void Normalize(const nnvm::NodeAttrs &attrs,
const OpContext &ctx,
const std::vector<TBlob> &inputs,
const std::vector<OpReqType> &req,
const std::vector<TBlob> &outputs) {
+ const NormalizeParam ¶m = nnvm::get<NormalizeParam>(attrs.parsed);
+ auto mean = param.mean;
+ auto std = param.std;
+
+ int nchannel = inputs[0].shape_[0];
+ NormalizeCheckParam(mean, std, nchannel);
+
+ int size = inputs[0].Size() / nchannel;
+ mshadow::Stream<xpu> *s = ctx.get_stream<xpu>();
+ MXNET_ASSIGN_REQ_SWITCH(req[0], Req, {
+ MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
+ mxnet_op::Kernel<normalize, xpu>::Launch(
+ s, inputs[0].Size(), outputs[0].dptr<DType>(), inputs[0].dptr<DType>(),
+ Req, nchannel, size, mean.begin(), std.begin());
+ });
+ });
+}
+
+template<typename xpu>
+static void NormalizeBackward(const nnvm::NodeAttrs &attrs,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+ const NormalizeParam ¶m = nnvm::get<NormalizeParam>(attrs.parsed);
+ int nchannel = inputs[0].shape_[0];
+
+ NormalizeCheckParam(param.mean, param.std, nchannel);
+
+ int size = inputs[0].Size() / nchannel;
+ mshadow::Stream<xpu> *s = ctx.get_stream<xpu>();
+ MXNET_ASSIGN_REQ_SWITCH(req[0], Req, {
+ MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
+ mxnet_op::Kernel<normalize, xpu>::Launch(
+ s, inputs[0].Size(), outputs[0].dptr<DType>(), inputs[0].dptr<DType>(),
+ Req, nchannel, size, nullptr, param.std.begin());
+ });
+ });
}
struct RandomBrightnessParam : public dmlc::Parameter<RandomBrightnessParam> {
diff --git a/src/operator/image/image_random.cc b/src/operator/image/image_random.cc
index 26fa843..e32a677 100644
--- a/src/operator/image/image_random.cc
+++ b/src/operator/image/image_random.cc
@@ -28,8 +28,6 @@
#include "../operator_common.h"
#include "../elemwise_op_common.h"
-
-
namespace mxnet {
namespace op {
@@ -46,6 +44,38 @@ NNVM_REGISTER_OP(_image_to_tensor)
.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
.add_argument("data", "NDArray-or-Symbol", "The input.");
+DMLC_REGISTER_PARAMETER(NormalizeParam);
+NNVM_REGISTER_OP(_image_normalize)
+.describe(R"code()code" ADD_FILELINE)
+.set_num_inputs(1)
+.set_num_outputs(1)
+.set_attr_parser(ParamParser<NormalizeParam>)
+.set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& attrs) {
+ return std::vector<ResourceRequest>{ResourceRequest::kRandom};
+})
+.set_attr<nnvm::FInferShape>("FInferShape", ElemwiseShape<1, 1>)
+.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
+.set_attr<nnvm::FInplaceOption>("FInplaceOption",
+[](const NodeAttrs& attrs){
+ return std::vector<std::pair<int, int> >{{0, 0}};
+})
+.set_attr<FCompute>("FCompute<cpu>", Normalize<cpu>)
+.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_image_backward_normalize" })
+.add_argument("data", "NDArray-or-Symbol", "The input.")
+.add_arguments(NormalizeParam::__FIELDS__());
+
+NNVM_REGISTER_OP(_image_backward_normalize)
+.describe(R"code()code" ADD_FILELINE)
+.set_num_inputs(1)
+.set_num_outputs(1)
+.set_attr_parser(ParamParser<NormalizeParam>)
+.set_attr<nnvm::TIsBackward>("TIsBackward", true)
+.set_attr<nnvm::FInplaceOption>("FInplaceOption",
+[](const NodeAttrs& attrs){
+ return std::vector<std::pair<int, int> >{{0, 0}};
+})
+.set_attr<FCompute>("FCompute<cpu>", NormalizeBackward<cpu>);
+
DMLC_REGISTER_PARAMETER(RandomBrightnessParam);
NNVM_REGISTER_OP(_image_random_brightness)
.describe(R"code()code" ADD_FILELINE)
--
To stop receiving notification emails like this one, please contact
"commits@mxnet.apache.org" <co...@mxnet.apache.org>.
[incubator-mxnet] 07/19: image to_tensor (#8691)
Posted by jx...@apache.org.
This is an automated email from the ASF dual-hosted git repository.
jxie pushed a commit to branch vision
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git
commit d5532bccdc6929091cbbea844a49b5ef60fc764f
Author: Yizhi Liu <ja...@gmail.com>
AuthorDate: Thu Nov 16 23:01:37 2017 -0800
image to_tensor (#8691)
---
src/operator/image/image_common.h | 7 +--
src/operator/image/image_random-inl.h | 92 +++++++++++++++++++++++++++--------
src/operator/image/image_random.cc | 21 ++++++--
src/operator/mxnet_op.h | 3 +-
4 files changed, 95 insertions(+), 28 deletions(-)
diff --git a/src/operator/image/image_common.h b/src/operator/image/image_common.h
index 7cf3f96..3b6b8e3 100644
--- a/src/operator/image/image_common.h
+++ b/src/operator/image/image_common.h
@@ -81,8 +81,9 @@ static cv::Mat mat_convert(TBlob input, int hight, int weight, int channel) {
}
return m;
}
-} // namespace op
-} // namespace mxnet
+} // namespace op
+} // namespace mxnet
-#endif // MXNET_OPERATOR_IMAGE_IMAGE_COMMON_H_
\ No newline at end of file
+#endif // MXNET_OPERATOR_IMAGE_IMAGE_COMMON_H_
+
diff --git a/src/operator/image/image_random-inl.h b/src/operator/image/image_random-inl.h
index 8a3acf6..c50ecb7 100644
--- a/src/operator/image/image_random-inl.h
+++ b/src/operator/image/image_random-inl.h
@@ -25,15 +25,14 @@
#ifndef MXNET_OPERATOR_IMAGE_IMAGE_RANDOM_INL_H_
#define MXNET_OPERATOR_IMAGE_IMAGE_RANDOM_INL_H_
-#include <vector>
#include <mxnet/base.h>
+#include <vector>
#include <opencv2/opencv.hpp>
#include <opencv2/core/mat.hpp>
#include "../mxnet_op.h"
#include "image_common.h"
#include "../../operator/operator_common.h"
-
namespace mxnet {
namespace op {
@@ -47,13 +46,66 @@ static void RandomFlip(const nnvm::NodeAttrs &attrs,
const std::vector<OpReqType> &req,
const std::vector<TBlob> &outputs) {
}
+
+inline bool ToTensorType(const nnvm::NodeAttrs& attrs,
+ std::vector<int> *in_attrs,
+ std::vector<int> *out_attrs) {
+ CHECK_EQ(in_attrs->size(), 1U);
+ CHECK_EQ(out_attrs->size(), 1U);
+ CHECK_EQ((*in_attrs)[0], mshadow::kUint8)
+ << "`to_tensor` only supports uint8 input";
+ TYPE_ASSIGN_CHECK(*out_attrs, 0, mshadow::kFloat32);
+ return (*in_attrs)[0] != -1;
+}
+
+inline bool ToTensorShape(const nnvm::NodeAttrs& attrs,
+ std::vector<TShape> *in_attrs,
+ std::vector<TShape> *out_attrs) {
+ CHECK_EQ(in_attrs->size(), 1U);
+ CHECK_EQ(out_attrs->size(), 1U);
+ TShape &shp = (*in_attrs)[0];
+ CHECK_EQ(shp.ndim(), 3U) << "`to_tensor` only supports 3 dimensions";
+ TShape ret(3);
+ ret[0] = shp[2];
+ ret[1] = shp[0];
+ ret[2] = shp[1];
+ SHAPE_ASSIGN_CHECK(*out_attrs, 0, ret);
+ return true;
+}
+
template<typename xpu>
static void ToTensor(const nnvm::NodeAttrs &attrs,
const OpContext &ctx,
const std::vector<TBlob> &inputs,
const std::vector<OpReqType> &req,
const std::vector<TBlob> &outputs) {
+ auto input = inputs[0];
+ auto output = outputs[0];
+
+ int height = input.shape_[0];
+ int weight = input.shape_[1];
+ int channel = input.shape_[2];
+
+ typedef float DstDType;
+ typedef uint8_t SrcDType;
+
+ CHECK_EQ(req[0], kWriteTo)
+ << "`to_tensor` does not support inplace";
+
+ mshadow::Stream<xpu> *s = ctx.get_stream<xpu>();
+ MXNET_ASSIGN_REQ_SWITCH(req[0], Req, {
+ auto input_3d = input.get<xpu, 3, SrcDType>(s);
+ auto output_3d = output.get<xpu, 3, DstDType>(s);
+ for (int h = 0; h < height; ++h) {
+ for (int w = 0; w < weight; ++w) {
+ for (int c = 0; c < channel; ++c) {
+ Assign(output_3d[c][h][w], Req, DstDType(input_3d[h][w][c] / 255.0));
+ }
+ }
+ }
+ });
}
+
template<typename xpu>
static void Normalize(const nnvm::NodeAttrs &attrs,
const OpContext &ctx,
@@ -81,20 +133,20 @@ static void RandomBrightness(const nnvm::NodeAttrs &attrs,
auto input = inputs[0];
auto output = outputs[0];
int channel = input.shape_[0];
- int hight = input.shape_[1];
+ int height = input.shape_[1];
int weight = input.shape_[2];
Stream<xpu> *s = ctx.get_stream<xpu>();
Random<xpu> *prnd = ctx.requested[kRandom].get_random<xpu, real_t>(s);
const RandomBrightnessParam ¶m = nnvm::get<RandomBrightnessParam>(attrs.parsed);
- float alpha_b = 1.0 + std::uniform_real_distribution<float>(-param.max_brightness, param.max_brightness)(prnd->GetRndEngine());
+ float alpha_b = 1.0 + std::uniform_real_distribution<float>(
+ -param.max_brightness, param.max_brightness)(prnd->GetRndEngine());
MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
MXNET_ASSIGN_REQ_SWITCH(req[0], Req, {
mxnet_op::Kernel<mxnet_op::op_with_req<mshadow::op::mul, Req>, xpu>::Launch(
s, inputs[0].Size(), outputs[0].dptr<DType>(), inputs[0].dptr<DType>(), DType(alpha_b));
});
});
-
}
struct RandomContrastParam : public dmlc::Parameter<RandomContrastParam> {
@@ -125,14 +177,15 @@ static void RandomContrast(const nnvm::NodeAttrs &attrs,
auto input = inputs[0];
auto output = outputs[0];
int channel = input.shape_[0];
- int hight = input.shape_[1];
+ int height = input.shape_[1];
int weight = input.shape_[2];
Stream<xpu> *s = ctx.get_stream<xpu>();
Random<xpu> *prnd = ctx.requested[kRandom].get_random<xpu, real_t>(s);
const RandomContrastParam ¶m = nnvm::get<RandomContrastParam>(attrs.parsed);
- float alpha_c = 1.0 + std::uniform_real_distribution<float>(-param.max_contrast, param.max_contrast)(prnd->GetRndEngine());
+ float alpha_c = 1.0 + std::uniform_real_distribution<float>(
+ -param.max_contrast, param.max_contrast)(prnd->GetRndEngine());
const float R2YF = 0.299f;
const float G2YF = 0.587f;
@@ -143,22 +196,21 @@ static void RandomContrast(const nnvm::NodeAttrs &attrs,
auto input_3d = input.get<xpu, 3, DType>(s);
DType sum = (DType)0.0;
for (int c = 0; c < channel; ++c) {
- for (int h = 0; h < hight; ++h) {
+ for (int h = 0; h < height; ++h) {
for (int w = 0; w < weight; ++w) {
sum += input_3d[c][h][w] * coeffs0[c];
}
}
}
- float gray_mean = sum / (float)(hight * weight);
+ float gray_mean = sum / static_cast<float>(height * weight);
float beta = (1 - alpha_c) * gray_mean;
MXNET_ASSIGN_REQ_SWITCH(req[0], Req, {
mxnet_op::Kernel<mxnet_op::op_with_req<mul_add, Req>, xpu>::Launch(
- s, inputs[0].Size(), outputs[0].dptr<DType>(), inputs[0].dptr<DType>(), DType(alpha_c), DType(beta));
+ s, inputs[0].Size(), outputs[0].dptr<DType>(),
+ inputs[0].dptr<DType>(), DType(alpha_c), DType(beta));
});
-
});
-
}
struct RandomSaturationParam : public dmlc::Parameter<RandomSaturationParam> {
@@ -180,12 +232,13 @@ static void RandomSaturation(const nnvm::NodeAttrs &attrs,
auto input = inputs[0];
auto output = outputs[0];
int channel = input.shape_[0];
- int hight = input.shape_[1];
+ int height = input.shape_[1];
int weight = input.shape_[2];
Stream<xpu> *s = ctx.get_stream<xpu>();
Random<xpu> *prnd = ctx.requested[kRandom].get_random<xpu, real_t>(s);
const RandomSaturationParam ¶m = nnvm::get<RandomSaturationParam>(attrs.parsed);
- float alpha_s = 1.0 + std::uniform_real_distribution<float>(-param.max_saturation, param.max_saturation)(prnd->GetRndEngine());
+ float alpha_s = 1.0 + std::uniform_real_distribution<float>(
+ -param.max_saturation, param.max_saturation)(prnd->GetRndEngine());
float alpha_o = 1 - alpha_s;
const float R2YF = 0.299f;
const float G2YF = 0.587f;
@@ -202,20 +255,19 @@ static void RandomSaturation(const nnvm::NodeAttrs &attrs,
Assign(output_3d, Req, input_3d)
break;
case 3:
- for (int h = 0; h < hight; ++h) {
+ for (int h = 0; h < height; ++h) {
for (int w = 0; w < weight; ++w) {
- float gray = input_3d[0][h][w] * R2YF + input_3d[1][h][w] * G2YF + input_3d[2][h][w] * B2YF;
+ float gray =
+ input_3d[0][h][w] * R2YF + input_3d[1][h][w] * G2YF + input_3d[2][h][w] * B2YF;
Assign(output_3d[0][h][w], Req, DType(gray * alpha_s + input_3d[0][h][w] * alpha_o))
}
}
break;
default:
LOG(FATAL) << "not support channel" << channel;
-
}
});
});
-
}
template<typename xpu>
@@ -245,7 +297,7 @@ static void RandomLighting(const nnvm::NodeAttrs &attrs,
-} // namespace op
-} // namespace mxnet
+} // namespace op
+} // namespace mxnet
#endif // MXNET_OPERATOR_IMAGE_IMAGE_RANDOM_INL_H_
diff --git a/src/operator/image/image_random.cc b/src/operator/image/image_random.cc
index 63f7904..26fa843 100644
--- a/src/operator/image/image_random.cc
+++ b/src/operator/image/image_random.cc
@@ -33,14 +33,27 @@
namespace mxnet {
namespace op {
+NNVM_REGISTER_OP(_image_to_tensor)
+.describe(R"code()code" ADD_FILELINE)
+.set_num_inputs(1)
+.set_num_outputs(1)
+.set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& attrs) {
+ return std::vector<ResourceRequest>{ResourceRequest::kRandom};
+})
+.set_attr<nnvm::FInferShape>("FInferShape", ToTensorShape)
+.set_attr<nnvm::FInferType>("FInferType", ToTensorType)
+.set_attr<FCompute>("FCompute<cpu>", ToTensor<cpu>)
+.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
+.add_argument("data", "NDArray-or-Symbol", "The input.");
+
DMLC_REGISTER_PARAMETER(RandomBrightnessParam);
NNVM_REGISTER_OP(_image_random_brightness)
.describe(R"code()code" ADD_FILELINE)
.set_num_inputs(1)
.set_num_outputs(1)
.set_attr_parser(ParamParser<RandomBrightnessParam>)
-.set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& attrs) {
- return std::vector<ResourceRequest>{ResourceRequest::kRandom};
+.set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& attrs) {
+ return std::vector<ResourceRequest>{ResourceRequest::kRandom};
})
.set_attr<nnvm::FInferShape>("FInferShape", ElemwiseShape<1, 1>)
.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
@@ -81,5 +94,5 @@ NNVM_REGISTER_OP(_image_random_saturation)
.add_argument("data", "NDArray-or-Symbol", "The input.")
.add_arguments(RandomSaturationParam::__FIELDS__());
-}
-}
+} // namespace op
+} // namespace mxnet
diff --git a/src/operator/mxnet_op.h b/src/operator/mxnet_op.h
index 43f3a55..5a36954 100644
--- a/src/operator/mxnet_op.h
+++ b/src/operator/mxnet_op.h
@@ -368,7 +368,8 @@ struct op_with_req {
/*! \brief input is tensor and two scalar value */
template<typename DType>
- MSHADOW_XINLINE static void Map(int i, DType *out, const DType *in, const DType value_1, const DType value_2) {
+ MSHADOW_XINLINE static void Map(int i, DType *out, const DType *in,
+ const DType value_1, const DType value_2) {
KERNEL_ASSIGN(out[i], req, OP::Map(in[i], value_1, value_2));
}
--
To stop receiving notification emails like this one, please contact
"commits@mxnet.apache.org" <co...@mxnet.apache.org>.
[incubator-mxnet] 12/19: Refactor image operators (#8761)
Posted by jx...@apache.org.
This is an automated email from the ASF dual-hosted git repository.
jxie pushed a commit to branch vision
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git
commit 05a5974ddee0b897b1346f0f87017820bd420d3e
Author: Eric Junyuan Xie <pi...@users.noreply.github.com>
AuthorDate: Tue Nov 21 17:43:21 2017 -0800
Refactor image operators (#8761)
* fix
* fix
* fix
* fix
* refactor
* fix
---
src/operator/image/image_aug_op.h | 70 --------
src/operator/image/image_common.h | 89 ----------
src/operator/image/image_random-inl.h | 314 ++++++++++++----------------------
src/operator/image/image_random.cc | 42 ++---
4 files changed, 134 insertions(+), 381 deletions(-)
diff --git a/src/operator/image/image_aug_op.h b/src/operator/image/image_aug_op.h
deleted file mode 100644
index 40315ec..0000000
--- a/src/operator/image/image_aug_op.h
+++ /dev/null
@@ -1,70 +0,0 @@
-/*
- * Licensed to the Apache Software Foundation (ASF) under one
- * or more contributor license agreements. See the NOTICE file
- * distributed with this work for additional information
- * regarding copyright ownership. The ASF licenses this file
- * to you under the Apache License, Version 2.0 (the
- * "License"); you may not use this file except in compliance
- * with the License. You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing,
- * software distributed under the License is distributed on an
- * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
- * KIND, either express or implied. See the License for the
- * specific language governing permissions and limitations
- * under the License.
- */
-
-#ifndef MXNET_OPERATOR_IMAGE_IMAGE_AUG_OP_H_
-#define MXNET_OPERATOR_IMAGE_IMAGE_AUG_OP_H_
-
-#include <mxnet/operator_util.h>
-#include <vector>
-#include <utility>
-#include <algorithm>
-#include "../mshadow_op.h"
-#include "../elemwise_op_common.h"
-#include "../mxnet_op.h"
-
-namespace mxnet {
-namespace op {
-
-struct NormalizeParam : public dmlc::Parameter<NormalizeParam> {
- nnvm::Tuple<float> mean, std;
- DMLC_DECLARE_PARAMETER(NormalizeParam) {
- DMLC_DECLARE_FIELD(mean).set_default(nnvm::Tuple<float>({0.f}))
- .describe("");
- DMLC_DECLARE_FIELD(std).set_default(nnvm::Tuple<float>({1.f}))
- .describe("");
- }
-};
-
-
-void NormalizeCompute(const nnvm::NodeAttrs& attrs,
- const OpContext& ctx,
- const std::vector<NDArray>& inputs,
- const std::vector<OpReqType>& req,
- const std::vector<NDArray>& outputs) {
- using namespace mxnet_op;
- const auto& params = dmlc::get<NormalizeParam>(attrs.parsed);
- CHECK_NE(req[0], kAddTo);
- MSHADOW_TYPE_SWITCH(inputs[0].type_flag_, DType, {
- auto num_channel = inputs[0].shape_[0];
- auto size = inputs[0].Size(1, inputs[0].ndim());
- nnvm::Tuple<DType> mean(params.mean.begin(), params.mean.end());
- nnvm::Tuple<DType> std(params.std.begin(), params.std.end());
- DType* src = inputs[0].dptr<DType>();
- DType* dst = outputs[0].dptr<DType>();
- for (int i = 0; i < num_channel; ++i) {
- for (int j = 0; j < size; ++j, ++out, ++src) {
- *out = (*src - mean[i]) / std[i];
- }
- }
- });
-}
-
-} // namespace op
-} // namespace mxnet
-#endif // MXNET_OPERATOR_IMAGE_IMAGE_AUG_OP_H_
diff --git a/src/operator/image/image_common.h b/src/operator/image/image_common.h
deleted file mode 100644
index 3b6b8e3..0000000
--- a/src/operator/image/image_common.h
+++ /dev/null
@@ -1,89 +0,0 @@
-/*
-* Licensed to the Apache Software Foundation (ASF) under one
-* or more contributor license agreements. See the NOTICE file
-* distributed with this work for additional information
-* regarding copyright ownership. The ASF licenses this file
-* to you under the Apache License, Version 2.0 (the
-* "License"); you may not use this file except in compliance
-* with the License. You may obtain a copy of the License at
-*
-* http://www.apache.org/licenses/LICENSE-2.0
-*
-* Unless required by applicable law or agreed to in writing,
-* software distributed under the License is distributed on an
-* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
-* KIND, either express or implied. See the License for the
-* specific language governing permissions and limitations
-* under the License.
-*/
-
-/*!
-* \file image_common.h
-* \brief
-* \author
-*/
-#ifndef MXNET_OPERATOR_IMAGE_IMAGE_COMMON_H_
-#define MXNET_OPERATOR_IMAGE_IMAGE_COMMON_H_
-
-#include <mxnet/base.h>
-
-namespace mxnet {
-namespace op {
-
-/**
-* @brief convert TBlob to cv::Mat
-* @param input @see TBlob
-* @param hight
-* @param weight
-* @param channel
-* @return
-*/
-static cv::Mat mat_convert(TBlob input, int hight, int weight, int channel) {
- cv::Mat m;
- switch (input.type_flag_) {
- case mshadow::kFloat32: {
- typedef float DType;
- m = cv::Mat(hight, weight, CV_MAKETYPE(CV_32F, channel), input.dptr<DType>());
- }
- break;
- case mshadow::kFloat64: {
- typedef double DType;
- m = cv::Mat(hight, weight, CV_MAKETYPE(CV_64F, channel), input.dptr<DType>());
- }
- break;
- case mshadow::kFloat16: {
- typedef mshadow::half::half_t DType;
- LOG(FATAL) << "not support type enum " << input.type_flag_;
- }
- break;
- case mshadow::kUint8: {
- typedef uint8_t DType;
- m = cv::Mat(hight, weight, CV_MAKETYPE(CV_8U, channel), input.dptr<DType>());
- }
- break;
- case mshadow::kInt8: {
- typedef int8_t DType;
- m = cv::Mat(hight, weight, CV_MAKETYPE(CV_8S, channel), input.dptr<DType>());
- }
- break;
- case mshadow::kInt32: {
- typedef int32_t DType;
- m = cv::Mat(hight, weight, CV_MAKETYPE(CV_32S, channel), input.dptr<DType>());
- }
- break;
- case mshadow::kInt64: {
- typedef int64_t DType;
- LOG(FATAL) << "not support type enum " << input.type_flag_;
- }
- break;
- default:
- LOG(FATAL) << "Unknown type enum " << input.type_flag_;
- }
- return m;
-}
-} // namespace op
-} // namespace mxnet
-
-
-#endif // MXNET_OPERATOR_IMAGE_IMAGE_COMMON_H_
-
diff --git a/src/operator/image/image_random-inl.h b/src/operator/image/image_random-inl.h
index 6f9cdc0..f823c8c 100644
--- a/src/operator/image/image_random-inl.h
+++ b/src/operator/image/image_random-inl.h
@@ -30,16 +30,11 @@
#include <opencv2/opencv.hpp>
#include <opencv2/core/mat.hpp>
#include "../mxnet_op.h"
-#include "image_common.h"
-#include "../../operator/operator_common.h"
+#include "../operator_common.h"
namespace mxnet {
namespace op {
-
-enum ImageRandomResource { kRandom };
-
-template<typename xpu>
static void RandomFlip(const nnvm::NodeAttrs &attrs,
const OpContext &ctx,
const std::vector<TBlob> &inputs,
@@ -73,37 +68,25 @@ inline bool ToTensorShape(const nnvm::NodeAttrs& attrs,
return true;
}
-template<typename xpu>
static void ToTensor(const nnvm::NodeAttrs &attrs,
const OpContext &ctx,
const std::vector<TBlob> &inputs,
const std::vector<OpReqType> &req,
const std::vector<TBlob> &outputs) {
- auto input = inputs[0];
- auto output = outputs[0];
-
- int height = input.shape_[0];
- int weight = input.shape_[1];
- int channel = input.shape_[2];
-
- typedef float DstDType;
- typedef uint8_t SrcDType;
-
CHECK_EQ(req[0], kWriteTo)
<< "`to_tensor` does not support inplace";
- mshadow::Stream<xpu> *s = ctx.get_stream<xpu>();
- MXNET_ASSIGN_REQ_SWITCH(req[0], Req, {
- auto input_3d = input.get<xpu, 3, SrcDType>(s);
- auto output_3d = output.get<xpu, 3, DstDType>(s);
- for (int h = 0; h < height; ++h) {
- for (int w = 0; w < weight; ++w) {
- for (int c = 0; c < channel; ++c) {
- Assign(output_3d[c][h][w], Req, DstDType(input_3d[h][w][c] / 255.0));
- }
- }
+ int length = inputs[0].shape_[0] * inputs[0].shape_[1];
+ int channel = inputs[0].shape_[2];
+
+ float* output = outputs[0].dptr<float>();
+ uint8_t* input = inputs[0].dptr<uint8_t>();
+
+ for (int l = 0; l < length; ++l) {
+ for (int c = 0; c < channel; ++c) {
+ output[c*length + l] = static_cast<float>(input[l*channel + c]) / 255.0f;
}
- });
+ }
}
struct NormalizeParam : public dmlc::Parameter<NormalizeParam> {
@@ -117,93 +100,47 @@ struct NormalizeParam : public dmlc::Parameter<NormalizeParam> {
}
};
-struct normalize {
- template<typename DType>
- MSHADOW_XINLINE static void Map(int i, DType *out, const DType *in,
- const OpReqType req,
- const int nchannel, const int size,
- const float *mean, const float *std) {
- int c = 0;
- switch (nchannel) {
- case 1:
- break;
- case 3:
- if (i < size) {
- c = 0;
- } else if (i < (size << 1)) {
- c = 1;
- } else {
- c = 2;
- }
- break;
- default:
- LOG(FATAL) << "not support channel" << nchannel;
- }
- float m = (mean ? mean[c] : 0);
- KERNEL_ASSIGN(out[i], req, static_cast<DType>((in[i] - m) / std[c]));
- }
-};
-static void NormalizeCheckParam(const nnvm::Tuple<float> &mean,
- const nnvm::Tuple<float> &std,
- const int nchannel) {
- CHECK(mean.ndim() == 1 || mean.ndim() == 3)
- << "Mean must be in dimension 1 or 3.";
- CHECK(std.ndim() == 1 || std.ndim() == 3)
- << "Standard deviations must be in dimension 1 or 3.";
- CHECK(nchannel == 1 || nchannel == 3) << "Image channel must be 1 or 3.";
- CHECK_EQ(mean.ndim(), nchannel)
- << "Mean dimension does not agree with image channel.";
- CHECK_EQ(std.ndim(), nchannel)
- << "Standard deviations dimension does not agree with image channel.";
- for (uint32_t c = 0; c < std.ndim(); ++c) {
- CHECK(std[c] > 0) << "Invalid standard deviation " << std[c];
- }
+inline bool NormalizeShape(const nnvm::NodeAttrs& attrs,
+ std::vector<TShape> *in_attrs,
+ std::vector<TShape> *out_attrs) {
+ const NormalizeParam ¶m = nnvm::get<NormalizeParam>(attrs.parsed);
+ const auto& dshape = (*in_attrs)[0];
+ if (!dshape.ndim()) return false;
+ CHECK_EQ(dshape.ndim(), 3)
+ << "Input must have 3 dimensions";
+
+ auto nchannels = dshape[0];
+ CHECK(param.mean.ndim() == 1 || param.mean.ndim() == nchannels)
+ << "mean must have either 1 or " << nchannels << " elements";
+ CHECK(param.std.ndim() == 1 || param.std.ndim() == nchannels)
+ << "std must have either 1 or " << nchannels << " elements";
+
+ SHAPE_ASSIGN_CHECK(*out_attrs, 0, dshape);
}
-template<typename xpu>
+
static void Normalize(const nnvm::NodeAttrs &attrs,
const OpContext &ctx,
const std::vector<TBlob> &inputs,
const std::vector<OpReqType> &req,
const std::vector<TBlob> &outputs) {
const NormalizeParam ¶m = nnvm::get<NormalizeParam>(attrs.parsed);
- auto mean = param.mean;
- auto std = param.std;
-
- int nchannel = inputs[0].shape_[0];
- NormalizeCheckParam(mean, std, nchannel);
-
- int size = inputs[0].Size() / nchannel;
- mshadow::Stream<xpu> *s = ctx.get_stream<xpu>();
- MXNET_ASSIGN_REQ_SWITCH(req[0], Req, {
- MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
- mxnet_op::Kernel<normalize, xpu>::Launch(
- s, inputs[0].Size(), outputs[0].dptr<DType>(), inputs[0].dptr<DType>(),
- Req, nchannel, size, mean.begin(), std.begin());
- });
- });
-}
-template<typename xpu>
-static void NormalizeBackward(const nnvm::NodeAttrs &attrs,
- const OpContext &ctx,
- const std::vector<TBlob> &inputs,
- const std::vector<OpReqType> &req,
- const std::vector<TBlob> &outputs) {
- const NormalizeParam ¶m = nnvm::get<NormalizeParam>(attrs.parsed);
- int nchannel = inputs[0].shape_[0];
-
- NormalizeCheckParam(param.mean, param.std, nchannel);
-
- int size = inputs[0].Size() / nchannel;
- mshadow::Stream<xpu> *s = ctx.get_stream<xpu>();
- MXNET_ASSIGN_REQ_SWITCH(req[0], Req, {
- MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
- mxnet_op::Kernel<normalize, xpu>::Launch(
- s, inputs[0].Size(), outputs[0].dptr<DType>(), inputs[0].dptr<DType>(),
- Req, nchannel, size, nullptr, param.std.begin());
- });
+ int nchannels = inputs[0].shape_[0];
+ int length = inputs[0].shape_[1] * inputs[0].shape_[2];
+
+ MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
+ DType* input = inputs[0].dptr<DType>();
+ DType* output = outputs[0].dptr<DType>();
+
+ for (int i = 0; i < nchannels; ++i) {
+ DType mean = param.mean[param.mean.ndim() > 1 ? i : 0];
+ DType std = param.std[param.std.ndim() > 1 ? i : 0];
+ for (int j = 0; j < length; ++j) {
+ output[i*length + j] = (input[i*length + j] - mean) / std;
+ }
+ }
});
}
@@ -211,99 +148,83 @@ struct RandomBrightnessParam : public dmlc::Parameter<RandomBrightnessParam> {
float max_brightness;
DMLC_DECLARE_PARAMETER(RandomBrightnessParam) {
DMLC_DECLARE_FIELD(max_brightness)
- .set_default(0.0)
+ .set_lower_bound(0.0)
.describe("Max Brightness.");
}
};
-template<typename xpu>
static void RandomBrightness(const nnvm::NodeAttrs &attrs,
const OpContext &ctx,
const std::vector<TBlob> &inputs,
const std::vector<OpReqType> &req,
const std::vector<TBlob> &outputs) {
using namespace mshadow;
- auto input = inputs[0];
- auto output = outputs[0];
- int channel = input.shape_[0];
- int height = input.shape_[1];
- int weight = input.shape_[2];
- Stream<xpu> *s = ctx.get_stream<xpu>();
- Random<xpu> *prnd = ctx.requested[kRandom].get_random<xpu, real_t>(s);
-
const RandomBrightnessParam ¶m = nnvm::get<RandomBrightnessParam>(attrs.parsed);
+
+ int length = inputs[0].Size();
+
+ uint8_t* output = outputs[0].dptr<uint8_t>();
+ uint8_t* input = inputs[0].dptr<uint8_t>();
+
+ Stream<cpu> *s = ctx.get_stream<cpu>();
+ Random<cpu> *prnd = ctx.requested[0].get_random<cpu, float>(s);
float alpha_b = 1.0 + std::uniform_real_distribution<float>(
- -param.max_brightness, param.max_brightness)(prnd->GetRndEngine());
- MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
- MXNET_ASSIGN_REQ_SWITCH(req[0], Req, {
- mxnet_op::Kernel<mxnet_op::op_with_req<mshadow::op::mul, Req>, xpu>::Launch(
- s, inputs[0].Size(), outputs[0].dptr<DType>(), inputs[0].dptr<DType>(), DType(alpha_b));
- });
- });
+ -param.max_brightness, param.max_brightness)(prnd->GetRndEngine());
+
+ for (int l = 0; l < length; ++l) {
+ float val = static_cast<float>(input[l]) * alpha_b;
+ val = std::min(std::max(val, 0.f), 255.f);
+ output[l] = static_cast<uint8_t>(val);
+ }
}
+
struct RandomContrastParam : public dmlc::Parameter<RandomContrastParam> {
float max_contrast;
DMLC_DECLARE_PARAMETER(RandomContrastParam) {
DMLC_DECLARE_FIELD(max_contrast)
- .set_default(0.0)
+ .set_lower_bound(0.0)
.describe("Max Contrast.");
}
};
-/*! \brief mul_add operator */
-struct mul_add {
- /*! \brief map a, b, c to result using defined operation */
- template<typename DType>
- MSHADOW_XINLINE static DType Map(DType a, DType b, DType c) {
- return a * b + c;
- }
-};
-template<typename xpu>
static void RandomContrast(const nnvm::NodeAttrs &attrs,
const OpContext &ctx,
const std::vector<TBlob> &inputs,
const std::vector<OpReqType> &req,
const std::vector<TBlob> &outputs) {
using namespace mshadow;
- auto input = inputs[0];
- auto output = outputs[0];
- int channel = input.shape_[0];
- int height = input.shape_[1];
- int weight = input.shape_[2];
- Stream<xpu> *s = ctx.get_stream<xpu>();
- Random<xpu> *prnd = ctx.requested[kRandom].get_random<xpu, real_t>(s);
+ static const float coef[] = { 0.299f, 0.587f, 0.114f };
+ const RandomContrastParam ¶m = nnvm::get<RandomContrastParam>(attrs.parsed);
+ int length = inputs[0].shape_[0] * inputs[0].shape_[1];
+ int nchannels = inputs[0].shape_[2];
- const RandomContrastParam ¶m = nnvm::get<RandomContrastParam>(attrs.parsed);
+ uint8_t* output = outputs[0].dptr<uint8_t>();
+ uint8_t* input = inputs[0].dptr<uint8_t>();
+
+ Stream<cpu> *s = ctx.get_stream<cpu>();
+ Random<cpu> *prnd = ctx.requested[0].get_random<cpu, real_t>(s);
float alpha_c = 1.0 + std::uniform_real_distribution<float>(
-param.max_contrast, param.max_contrast)(prnd->GetRndEngine());
- const float R2YF = 0.299f;
- const float G2YF = 0.587f;
- const float B2YF = 0.114f;
- static const float coeffs0[] = { R2YF, G2YF, B2YF };
-
- MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
- auto input_3d = input.get<xpu, 3, DType>(s);
- DType sum = (DType)0.0;
- for (int c = 0; c < channel; ++c) {
- for (int h = 0; h < height; ++h) {
- for (int w = 0; w < weight; ++w) {
- sum += input_3d[c][h][w] * coeffs0[c];
- }
- }
+ float sum = 0.f;
+ if (nchannels > 1) {
+ for (int l = 0; l < length; ++l) {
+ for (int c = 0; c < nchannels; ++c) sum += input[l*nchannels + c] * coef[c];
}
- float gray_mean = sum / static_cast<float>(height * weight);
- float beta = (1 - alpha_c) * gray_mean;
-
- MXNET_ASSIGN_REQ_SWITCH(req[0], Req, {
- mxnet_op::Kernel<mxnet_op::op_with_req<mul_add, Req>, xpu>::Launch(
- s, inputs[0].Size(), outputs[0].dptr<DType>(),
- inputs[0].dptr<DType>(), DType(alpha_c), DType(beta));
- });
- });
+ } else {
+ for (int l = 0; l < length; ++l) sum += input[l];
+ }
+ float gray_mean = sum / static_cast<float>(length);
+ float beta = (1 - alpha_c) * gray_mean;
+
+ for (int l = 0; l < length * nchannels; ++l) {
+ float val = input[l] * alpha_c + beta;
+ val = std::min(std::max(val, 0.f), 255.f);
+ output[l] = static_cast<uint8_t>(val);
+ }
}
struct RandomSaturationParam : public dmlc::Parameter<RandomSaturationParam> {
@@ -315,55 +236,46 @@ struct RandomSaturationParam : public dmlc::Parameter<RandomSaturationParam> {
}
};
-template<typename xpu>
static void RandomSaturation(const nnvm::NodeAttrs &attrs,
const OpContext &ctx,
const std::vector<TBlob> &inputs,
const std::vector<OpReqType> &req,
const std::vector<TBlob> &outputs) {
using namespace mshadow;
- auto input = inputs[0];
- auto output = outputs[0];
- int channel = input.shape_[0];
- int height = input.shape_[1];
- int weight = input.shape_[2];
- Stream<xpu> *s = ctx.get_stream<xpu>();
- Random<xpu> *prnd = ctx.requested[kRandom].get_random<xpu, real_t>(s);
const RandomSaturationParam ¶m = nnvm::get<RandomSaturationParam>(attrs.parsed);
- float alpha_s = 1.0 + std::uniform_real_distribution<float>(
+ static const float coef[] = { 0.299f, 0.587f, 0.114f };
+
+ int length = inputs[0].shape_[0] * inputs[0].shape_[1];
+ int nchannels = inputs[0].shape_[2];
+
+ uint8_t* output = outputs[0].dptr<uint8_t>();
+ uint8_t* input = inputs[0].dptr<uint8_t>();
+
+ Stream<cpu> *s = ctx.get_stream<cpu>();
+ Random<cpu> *prnd = ctx.requested[0].get_random<cpu, real_t>(s);
+ float alpha_s = 1.f + std::uniform_real_distribution<float>(
-param.max_saturation, param.max_saturation)(prnd->GetRndEngine());
- float alpha_o = 1 - alpha_s;
- const float R2YF = 0.299f;
- const float G2YF = 0.587f;
- const float B2YF = 0.114f;
- static const float coeffs0[] = { R2YF, G2YF, B2YF };
+ float alpha_o = 1.f - alpha_s;
+ if (nchannels == 1) {
+ for (int l = 0; l < length * nchannels; ++l) output[l] = input[l];
+ return;
+ }
- MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
- MXNET_ASSIGN_REQ_SWITCH(req[0], Req, {
- auto input_3d = input.get<xpu, 3, DType>(s);
- auto output_3d = output.get<xpu, 3, DType>(s);
- switch (channel) {
- case 1:
- Assign(output_3d, Req, input_3d)
- break;
- case 3:
- for (int h = 0; h < height; ++h) {
- for (int w = 0; w < weight; ++w) {
- float gray =
- input_3d[0][h][w] * R2YF + input_3d[1][h][w] * G2YF + input_3d[2][h][w] * B2YF;
- Assign(output_3d[0][h][w], Req, DType(gray * alpha_s + input_3d[0][h][w] * alpha_o))
- }
- }
- break;
- default:
- LOG(FATAL) << "not support channel" << channel;
- }
- });
- });
+ for (int l = 0; l < length; ++l) {
+ float gray = 0.f;
+ for (int c = 0; c < nchannels; ++c) {
+ gray = input[l*nchannels + c] * coef[c];
+ }
+ gray *= alpha_o;
+ for (int c = 0; c < nchannels; ++c) {
+ float val = gray + input[l*nchannels + c] * alpha_s;
+ val = std::min(std::max(val, 0.f), 255.f);
+ output[l*nchannels + c] = static_cast<uint8_t>(val);
+ }
+ }
}
-template<typename xpu>
static void RandomHue(const nnvm::NodeAttrs &attrs,
const OpContext &ctx,
const std::vector<TBlob> &inputs,
@@ -371,7 +283,6 @@ static void RandomHue(const nnvm::NodeAttrs &attrs,
const std::vector<TBlob> &outputs) {
}
-template<typename xpu>
static void RandomColorJitter(const nnvm::NodeAttrs &attrs,
const OpContext &ctx,
const std::vector<TBlob> &inputs,
@@ -379,7 +290,6 @@ static void RandomColorJitter(const nnvm::NodeAttrs &attrs,
const std::vector<TBlob> &outputs) {
}
-template<typename xpu>
static void RandomLighting(const nnvm::NodeAttrs &attrs,
const OpContext &ctx,
const std::vector<TBlob> &inputs,
diff --git a/src/operator/image/image_random.cc b/src/operator/image/image_random.cc
index e32a677..7ff7328 100644
--- a/src/operator/image/image_random.cc
+++ b/src/operator/image/image_random.cc
@@ -40,10 +40,11 @@ NNVM_REGISTER_OP(_image_to_tensor)
})
.set_attr<nnvm::FInferShape>("FInferShape", ToTensorShape)
.set_attr<nnvm::FInferType>("FInferType", ToTensorType)
-.set_attr<FCompute>("FCompute<cpu>", ToTensor<cpu>)
+.set_attr<FCompute>("FCompute<cpu>", ToTensor)
.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
.add_argument("data", "NDArray-or-Symbol", "The input.");
+
DMLC_REGISTER_PARAMETER(NormalizeParam);
NNVM_REGISTER_OP(_image_normalize)
.describe(R"code()code" ADD_FILELINE)
@@ -56,25 +57,14 @@ NNVM_REGISTER_OP(_image_normalize)
.set_attr<nnvm::FInferShape>("FInferShape", ElemwiseShape<1, 1>)
.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
.set_attr<nnvm::FInplaceOption>("FInplaceOption",
-[](const NodeAttrs& attrs){
- return std::vector<std::pair<int, int> >{{0, 0}};
-})
-.set_attr<FCompute>("FCompute<cpu>", Normalize<cpu>)
-.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_image_backward_normalize" })
+ [](const NodeAttrs& attrs){
+ return std::vector<std::pair<int, int> >{{0, 0}};
+ })
+.set_attr<FCompute>("FCompute<cpu>", Normalize)
+.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
.add_argument("data", "NDArray-or-Symbol", "The input.")
.add_arguments(NormalizeParam::__FIELDS__());
-NNVM_REGISTER_OP(_image_backward_normalize)
-.describe(R"code()code" ADD_FILELINE)
-.set_num_inputs(1)
-.set_num_outputs(1)
-.set_attr_parser(ParamParser<NormalizeParam>)
-.set_attr<nnvm::TIsBackward>("TIsBackward", true)
-.set_attr<nnvm::FInplaceOption>("FInplaceOption",
-[](const NodeAttrs& attrs){
- return std::vector<std::pair<int, int> >{{0, 0}};
-})
-.set_attr<FCompute>("FCompute<cpu>", NormalizeBackward<cpu>);
DMLC_REGISTER_PARAMETER(RandomBrightnessParam);
NNVM_REGISTER_OP(_image_random_brightness)
@@ -87,7 +77,11 @@ NNVM_REGISTER_OP(_image_random_brightness)
})
.set_attr<nnvm::FInferShape>("FInferShape", ElemwiseShape<1, 1>)
.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
-.set_attr<FCompute>("FCompute<cpu>", RandomBrightness<cpu>)
+.set_attr<nnvm::FInplaceOption>("FInplaceOption",
+ [](const NodeAttrs& attrs){
+ return std::vector<std::pair<int, int> >{{0, 0}};
+ })
+.set_attr<FCompute>("FCompute<cpu>", RandomBrightness)
.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
.add_argument("data", "NDArray-or-Symbol", "The input.")
.add_arguments(RandomBrightnessParam::__FIELDS__());
@@ -103,7 +97,11 @@ NNVM_REGISTER_OP(_image_random_contrast)
})
.set_attr<nnvm::FInferShape>("FInferShape", ElemwiseShape<1, 1>)
.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
-.set_attr<FCompute>("FCompute<cpu>", RandomContrast<cpu>)
+.set_attr<nnvm::FInplaceOption>("FInplaceOption",
+ [](const NodeAttrs& attrs){
+ return std::vector<std::pair<int, int> >{{0, 0}};
+ })
+.set_attr<FCompute>("FCompute<cpu>", RandomContrast)
.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
.add_argument("data", "NDArray-or-Symbol", "The input.")
.add_arguments(RandomContrastParam::__FIELDS__());
@@ -119,7 +117,11 @@ NNVM_REGISTER_OP(_image_random_saturation)
})
.set_attr<nnvm::FInferShape>("FInferShape", ElemwiseShape<1, 1>)
.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
-.set_attr<FCompute>("FCompute<cpu>", RandomSaturation<cpu>)
+.set_attr<nnvm::FInplaceOption>("FInplaceOption",
+ [](const NodeAttrs& attrs){
+ return std::vector<std::pair<int, int> >{{0, 0}};
+ })
+.set_attr<FCompute>("FCompute<cpu>", RandomSaturation)
.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
.add_argument("data", "NDArray-or-Symbol", "The input.")
.add_arguments(RandomSaturationParam::__FIELDS__());
--
To stop receiving notification emails like this one, please contact
"commits@mxnet.apache.org" <co...@mxnet.apache.org>.
[incubator-mxnet] 09/19: fix (#8736)
Posted by jx...@apache.org.
This is an automated email from the ASF dual-hosted git repository.
jxie pushed a commit to branch vision
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git
commit 7996da927ee269fb157531685e5affa2fd608cc7
Author: Eric Junyuan Xie <pi...@users.noreply.github.com>
AuthorDate: Mon Nov 20 22:06:32 2017 -0800
fix (#8736)
---
python/mxnet/gluon/data/dataset.py | 2 +-
python/mxnet/gluon/data/vision/datasets.py | 6 +++---
python/mxnet/gluon/data/vision/transforms.py | 2 +-
3 files changed, 5 insertions(+), 5 deletions(-)
diff --git a/python/mxnet/gluon/data/dataset.py b/python/mxnet/gluon/data/dataset.py
index 35d4c5c..f7ab395 100644
--- a/python/mxnet/gluon/data/dataset.py
+++ b/python/mxnet/gluon/data/dataset.py
@@ -18,7 +18,7 @@
# coding: utf-8
# pylint: disable=
"""Dataset container."""
-__all__ = ['Dataset', 'SimpleDataset', 'ArrayDataset', 'LabeledDataset',
+__all__ = ['Dataset', 'SimpleDataset', 'ArrayDataset',
'RecordFileDataset']
import os
diff --git a/python/mxnet/gluon/data/vision/datasets.py b/python/mxnet/gluon/data/vision/datasets.py
index cb40af5..4ddc2e3 100644
--- a/python/mxnet/gluon/data/vision/datasets.py
+++ b/python/mxnet/gluon/data/vision/datasets.py
@@ -28,9 +28,9 @@ import struct
import warnings
import numpy as np
-from . import dataset
-from ..utils import download, check_sha1
-from ... import nd, image, recordio
+from .. import dataset
+from ...utils import download, check_sha1
+from .... import nd, image, recordio
apache_repo_url = 'https://apache-mxnet.s3-accelerate.dualstack.amazonaws.com/'
diff --git a/python/mxnet/gluon/data/vision/transforms.py b/python/mxnet/gluon/data/vision/transforms.py
index fa7c0f2..e1deef6 100644
--- a/python/mxnet/gluon/data/vision/transforms.py
+++ b/python/mxnet/gluon/data/vision/transforms.py
@@ -58,7 +58,7 @@ class ToTensor(HybridBlock):
super(ToTensor, self).__init__()
def hybrid_forward(self, F, x):
- return F.cast(x, 'float32').transpose((2, 0, 1))
+ return F.image.to_tensor(x)
class Normalize(HybridBlock):
--
To stop receiving notification emails like this one, please contact
"commits@mxnet.apache.org" <co...@mxnet.apache.org>.
[incubator-mxnet] 06/19: Add Gluon data transform (#8672)
Posted by jx...@apache.org.
This is an automated email from the ASF dual-hosted git repository.
jxie pushed a commit to branch vision
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git
commit 80f323a0d0e807e52c3ec46cb8cba9b4569e6437
Author: Eric Junyuan Xie <pi...@users.noreply.github.com>
AuthorDate: Wed Nov 15 15:16:51 2017 -0800
Add Gluon data transform (#8672)
* fix
* fix
* fix
---
python/mxnet/gluon/data/dataset.py | 43 +++++-
python/mxnet/gluon/data/vision/__init__.py | 22 +++
.../gluon/data/{vision.py => vision/datasets.py} | 0
python/mxnet/gluon/data/vision/transforms.py | 153 +++++++++++++++++++++
src/operator/image/image_aug_op.h | 70 ++++++++++
src/operator/image/image_random.cc | 4 +-
6 files changed, 289 insertions(+), 3 deletions(-)
diff --git a/python/mxnet/gluon/data/dataset.py b/python/mxnet/gluon/data/dataset.py
index 2c46f1e..35d4c5c 100644
--- a/python/mxnet/gluon/data/dataset.py
+++ b/python/mxnet/gluon/data/dataset.py
@@ -18,12 +18,14 @@
# coding: utf-8
# pylint: disable=
"""Dataset container."""
-__all__ = ['Dataset', 'ArrayDataset', 'RecordFileDataset']
+__all__ = ['Dataset', 'SimpleDataset', 'ArrayDataset', 'LabeledDataset',
+ 'RecordFileDataset']
import os
from ... import recordio, ndarray
+
class Dataset(object):
"""Abstract dataset class. All datasets should have this interface.
@@ -38,6 +40,45 @@ class Dataset(object):
def __len__(self):
raise NotImplementedError
+ def transform(self, fn, lazy=True):
+ trans = _LazyTransformDataset(self, fn)
+ if lazy:
+ return trans
+ return SimpleDataset([i for i in trans])
+
+ def transform_first(self, fn, lazy=True):
+ def base_fn(x, *args):
+ if args:
+ return (fn(x),) + args
+ return fn(x)
+ return self.transform(base_fn, lazy)
+
+
+class SimpleDataset(Dataset):
+ def __init__(self, data):
+ self._data = data
+
+ def __len__(self):
+ return len(self._data)
+
+ def __getitem__(self, idx):
+ return self._data[idx]
+
+
+class _LazyTransformDataset(Dataset):
+ def __init__(self, data, fn):
+ self._data = data
+ self._fn = fn
+
+ def __len__(self):
+ return len(self._data)
+
+ def __getitem__(self, idx):
+ item = self._data[idx]
+ if isinstance(item, tuple):
+ return self._fn(*item)
+ return self._fn(item)
+
class ArrayDataset(Dataset):
"""A dataset of multiple arrays.
diff --git a/python/mxnet/gluon/data/vision/__init__.py b/python/mxnet/gluon/data/vision/__init__.py
new file mode 100644
index 0000000..8837984
--- /dev/null
+++ b/python/mxnet/gluon/data/vision/__init__.py
@@ -0,0 +1,22 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements. See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership. The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License. You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied. See the License for the
+# specific language governing permissions and limitations
+# under the License.
+
+# coding: utf-8
+
+from .datasets import *
+
+from . import transforms
diff --git a/python/mxnet/gluon/data/vision.py b/python/mxnet/gluon/data/vision/datasets.py
similarity index 100%
rename from python/mxnet/gluon/data/vision.py
rename to python/mxnet/gluon/data/vision/datasets.py
diff --git a/python/mxnet/gluon/data/vision/transforms.py b/python/mxnet/gluon/data/vision/transforms.py
new file mode 100644
index 0000000..fa7c0f2
--- /dev/null
+++ b/python/mxnet/gluon/data/vision/transforms.py
@@ -0,0 +1,153 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements. See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership. The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License. You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied. See the License for the
+# specific language governing permissions and limitations
+# under the License.
+
+# coding: utf-8
+
+from .. import dataset
+from ...block import Block, HybridBlock
+from ...nn import Sequential, HybridSequential
+from .... import ndarray, initializer
+
+
+class Compose(Sequential):
+ def __init__(self, transforms):
+ super(Compose, self).__init__()
+ transforms.append(None)
+ hybrid = []
+ for i in transforms:
+ if isinstance(i, HybridBlock):
+ hybrid.append(i)
+ continue
+ elif len(hybrid) == 1:
+ self.register_child(hybrid[0])
+ elif len(hybrid) > 1:
+ hblock = HybridSequential()
+ for j in hybrid:
+ hblock.add(j)
+ self.register_child(hblock)
+ if i is not None:
+ self.register_child(i)
+ self.hybridize()
+
+
+class Cast(HybridBlock):
+ def __init__(self, dtype='float32'):
+ super(Cast, self).__init__()
+ self._dtype = dtype
+
+ def hybrid_forward(self, F, x):
+ return F.cast(x, self._dtype)
+
+
+class ToTensor(HybridBlock):
+ def __init__(self):
+ super(ToTensor, self).__init__()
+
+ def hybrid_forward(self, F, x):
+ return F.cast(x, 'float32').transpose((2, 0, 1))
+
+
+class Normalize(HybridBlock):
+ def __init__(self, mean, std):
+ super(Normalize, self).__init__()
+ self._mean = mean
+ self._std = std
+
+ def hybrid_forward(self, F, x):
+ return F.image.normalize(x, self._mean, self._std)
+
+
+class RandomResizedCrop(HybridBlock):
+ def __init__(self, size, area=(0.08, 1.0), ratio=(3.0/4.0, 4.0/3.0),
+ interpolation=2):
+ super(RandomResizedCrop, self).__init__()
+ self._args = (size, area, ratio, interpolation)
+
+ def hybrid_forward(self, F, x):
+ return F.image.random_resized_crop(x, *self._args)
+
+
+class CenterCrop(HybridBlock):
+ def __init__(self, size):
+ super(CenterCrop, self).__init__()
+ self._size = size
+
+ def hybrid_forward(self, F, x):
+ return F.image.center_crop(x, size)
+
+
+class Resize(HybridBlock):
+ def __init__(self, size, interpolation=2):
+ super(Resize, self).__init__()
+ self._args = (size, interpolation)
+
+ def hybrid_forward(self, F, x):
+ return F.image.resize(x, *self._args)
+
+
+class RandomFlip(HybridBlock):
+ def __init__(self, axis=1):
+ super(RandomFlip, self).__init__()
+ self._axis = axis
+
+ def hybrid_forward(self, F, x):
+ return F.image.random_flip(x, self._axis)
+
+
+class RandomBrightness(HybridBlock):
+ def __init__(self, max_brightness):
+ super(RandomBrightness, self).__init__()
+ self._max_brightness = max_brightness
+
+ def hybrid_forward(self, F, x):
+ return F.image.random_brightness(x, self._max_brightness)
+
+
+class RandomContrast(HybridBlock):
+ def __init__(self, max_contrast):
+ super(RandomContrast, self).__init__()
+ self._max_contrast = max_contrast
+
+ def hybrid_forward(self, F, x):
+ return F.image.random_contrast(x, self._max_contrast)
+
+
+class RandomSaturation(HybridBlock):
+ def __init__(self, max_saturation):
+ super(RandomSaturation, self).__init__()
+ self._max_saturation = max_saturation
+
+ def hybrid_forward(self, F, x):
+ return F.image.random_saturation(x, self._max_saturation)
+
+
+class RandomHue(HybridBlock):
+ def __init__(self, max_hue):
+ super(RandomHue, self).__init__()
+ self._max_hue = max_hue
+
+ def hybrid_forward(self, F, x):
+ return F.image.random_hue(x, self._max_hue)
+
+
+class RandomColorJitter(HybridBlock):
+ def __init__(self, max_brightness=0, max_contrast=0, max_saturation=0, max_hue=0):
+ super(RandomColorJitter, self).__init__()
+ self._args = (max_brightness, max_contrast, max_saturation, max_hue)
+
+ def hybrid_forward(self, F, x):
+ return F.image.random_color_jitter(x, *self._args)
diff --git a/src/operator/image/image_aug_op.h b/src/operator/image/image_aug_op.h
new file mode 100644
index 0000000..40315ec
--- /dev/null
+++ b/src/operator/image/image_aug_op.h
@@ -0,0 +1,70 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+#ifndef MXNET_OPERATOR_IMAGE_IMAGE_AUG_OP_H_
+#define MXNET_OPERATOR_IMAGE_IMAGE_AUG_OP_H_
+
+#include <mxnet/operator_util.h>
+#include <vector>
+#include <utility>
+#include <algorithm>
+#include "../mshadow_op.h"
+#include "../elemwise_op_common.h"
+#include "../mxnet_op.h"
+
+namespace mxnet {
+namespace op {
+
+struct NormalizeParam : public dmlc::Parameter<NormalizeParam> {
+ nnvm::Tuple<float> mean, std;
+ DMLC_DECLARE_PARAMETER(NormalizeParam) {
+ DMLC_DECLARE_FIELD(mean).set_default(nnvm::Tuple<float>({0.f}))
+ .describe("");
+ DMLC_DECLARE_FIELD(std).set_default(nnvm::Tuple<float>({1.f}))
+ .describe("");
+ }
+};
+
+
+void NormalizeCompute(const nnvm::NodeAttrs& attrs,
+ const OpContext& ctx,
+ const std::vector<NDArray>& inputs,
+ const std::vector<OpReqType>& req,
+ const std::vector<NDArray>& outputs) {
+ using namespace mxnet_op;
+ const auto& params = dmlc::get<NormalizeParam>(attrs.parsed);
+ CHECK_NE(req[0], kAddTo);
+ MSHADOW_TYPE_SWITCH(inputs[0].type_flag_, DType, {
+ auto num_channel = inputs[0].shape_[0];
+ auto size = inputs[0].Size(1, inputs[0].ndim());
+ nnvm::Tuple<DType> mean(params.mean.begin(), params.mean.end());
+ nnvm::Tuple<DType> std(params.std.begin(), params.std.end());
+ DType* src = inputs[0].dptr<DType>();
+ DType* dst = outputs[0].dptr<DType>();
+ for (int i = 0; i < num_channel; ++i) {
+ for (int j = 0; j < size; ++j, ++out, ++src) {
+ *out = (*src - mean[i]) / std[i];
+ }
+ }
+ });
+}
+
+} // namespace op
+} // namespace mxnet
+#endif // MXNET_OPERATOR_IMAGE_IMAGE_AUG_OP_H_
diff --git a/src/operator/image/image_random.cc b/src/operator/image/image_random.cc
index 83abc17..63f7904 100644
--- a/src/operator/image/image_random.cc
+++ b/src/operator/image/image_random.cc
@@ -25,8 +25,8 @@
#include <mxnet/base.h>
#include "./image_random-inl.h"
-#include "../../operator/operator_common.h"
-#include "../../operator/elemwise_op_common.h"
+#include "../operator_common.h"
+#include "../elemwise_op_common.h"
--
To stop receiving notification emails like this one, please contact
"commits@mxnet.apache.org" <co...@mxnet.apache.org>.
[incubator-mxnet] 05/19: fix image_random compile (#8665)
Posted by jx...@apache.org.
This is an automated email from the ASF dual-hosted git repository.
jxie pushed a commit to branch vision
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git
commit 383f4903fa26bb7003bb9e30d79fb7c08647c5f0
Author: Yizhi Liu <ja...@gmail.com>
AuthorDate: Wed Nov 15 10:40:34 2017 -0800
fix image_random compile (#8665)
---
src/operator/image/image_random-inl.h | 2 +-
src/operator/image/image_random.cc | 4 ++--
2 files changed, 3 insertions(+), 3 deletions(-)
diff --git a/src/operator/image/image_random-inl.h b/src/operator/image/image_random-inl.h
index fa1d6dc..8a3acf6 100644
--- a/src/operator/image/image_random-inl.h
+++ b/src/operator/image/image_random-inl.h
@@ -31,7 +31,7 @@
#include <opencv2/core/mat.hpp>
#include "../mxnet_op.h"
#include "image_common.h"
-#include "operator/operator_common.h"
+#include "../../operator/operator_common.h"
namespace mxnet {
diff --git a/src/operator/image/image_random.cc b/src/operator/image/image_random.cc
index 32648bb..83abc17 100644
--- a/src/operator/image/image_random.cc
+++ b/src/operator/image/image_random.cc
@@ -25,8 +25,8 @@
#include <mxnet/base.h>
#include "./image_random-inl.h"
-#include "operator/operator_common.h"
-#include "operator/elemwise_op_common.h"
+#include "../../operator/operator_common.h"
+#include "../../operator/elemwise_op_common.h"
--
To stop receiving notification emails like this one, please contact
"commits@mxnet.apache.org" <co...@mxnet.apache.org>.
[incubator-mxnet] 17/19: fix (#8857)
Posted by jx...@apache.org.
This is an automated email from the ASF dual-hosted git repository.
jxie pushed a commit to branch vision
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git
commit 9a7e3e6556d40ba441e3b26b282954291c9d76c2
Author: Eric Junyuan Xie <pi...@users.noreply.github.com>
AuthorDate: Tue Nov 28 17:28:58 2017 -0800
fix (#8857)
use macro for registration
---
src/operator/image/image_random-inl.h | 22 +++++++
src/operator/image/image_random.cc | 111 ++++------------------------------
2 files changed, 34 insertions(+), 99 deletions(-)
diff --git a/src/operator/image/image_random-inl.h b/src/operator/image/image_random-inl.h
index 9d10a30..cbc7f40 100644
--- a/src/operator/image/image_random-inl.h
+++ b/src/operator/image/image_random-inl.h
@@ -658,6 +658,28 @@ void RandomLighting(const nnvm::NodeAttrs &attrs,
AdjustLightingImpl({alpha_r, alpha_g, alpha_b}, ctx, inputs, req, outputs);
}
+
+#define MXNET_REGISTER_IMAGE_AUG_OP(name) \
+ NNVM_REGISTER_OP(name) \
+ .set_num_inputs(1) \
+ .set_num_outputs(1) \
+ .set_attr<nnvm::FInplaceOption>("FInplaceOption", \
+ [](const NodeAttrs& attrs){ \
+ return std::vector<std::pair<int, int> >{{0, 0}}; \
+ }) \
+ .set_attr<nnvm::FInferShape>("FInferShape", ImageShape) \
+ .set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>) \
+ .set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" }) \
+ .add_argument("data", "NDArray-or-Symbol", "The input.")
+
+
+#define MXNET_REGISTER_IMAGE_RND_AUG_OP(name) \
+ MXNET_REGISTER_IMAGE_AUG_OP(name) \
+ .set_attr<FResourceRequest>("FResourceRequest", \
+ [](const NodeAttrs& attrs) { \
+ return std::vector<ResourceRequest>{ResourceRequest::kRandom}; \
+ })
+
} // namespace image
} // namespace op
} // namespace mxnet
diff --git a/src/operator/image/image_random.cc b/src/operator/image/image_random.cc
index 5a21bf8..481dfce 100644
--- a/src/operator/image/image_random.cc
+++ b/src/operator/image/image_random.cc
@@ -66,144 +66,57 @@ NNVM_REGISTER_OP(_image_normalize)
.add_arguments(NormalizeParam::__FIELDS__());
-NNVM_REGISTER_OP(_image_random_horizontal_flip)
+MXNET_REGISTER_IMAGE_RND_AUG_OP(_image_random_horizontal_flip)
.describe(R"code()code" ADD_FILELINE)
-.set_num_inputs(1)
-.set_num_outputs(1)
-.set_attr<nnvm::FInferShape>("FInferShape", ImageShape)
-.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
-.set_attr<nnvm::FInplaceOption>("FInplaceOption",
- [](const NodeAttrs& attrs){
- return std::vector<std::pair<int, int> >{{0, 0}};
- })
-.set_attr<FCompute>("FCompute<cpu>", RandomHorizontalFlip)
-.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
-.add_argument("data", "NDArray-or-Symbol", "The input.");
+.set_attr<FCompute>("FCompute<cpu>", RandomHorizontalFlip);
-NNVM_REGISTER_OP(_image_random_brightness)
+MXNET_REGISTER_IMAGE_RND_AUG_OP(_image_random_brightness)
.describe(R"code()code" ADD_FILELINE)
-.set_num_inputs(1)
-.set_num_outputs(1)
.set_attr_parser(ParamParser<RandomEnhanceParam>)
-.set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& attrs) {
- return std::vector<ResourceRequest>{ResourceRequest::kRandom};
-})
-.set_attr<nnvm::FInferShape>("FInferShape", ImageShape)
-.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
-.set_attr<nnvm::FInplaceOption>("FInplaceOption",
- [](const NodeAttrs& attrs){
- return std::vector<std::pair<int, int> >{{0, 0}};
- })
.set_attr<FCompute>("FCompute<cpu>", RandomBrightness)
-.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
-.add_argument("data", "NDArray-or-Symbol", "The input.")
.add_arguments(RandomEnhanceParam::__FIELDS__());
-NNVM_REGISTER_OP(_image_random_contrast)
+MXNET_REGISTER_IMAGE_RND_AUG_OP(_image_random_contrast)
.describe(R"code()code" ADD_FILELINE)
-.set_num_inputs(1)
-.set_num_outputs(1)
.set_attr_parser(ParamParser<RandomEnhanceParam>)
-.set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& attrs) {
- return std::vector<ResourceRequest>{ResourceRequest::kRandom};
-})
-.set_attr<nnvm::FInferShape>("FInferShape", ImageShape)
-.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
-.set_attr<nnvm::FInplaceOption>("FInplaceOption",
- [](const NodeAttrs& attrs){
- return std::vector<std::pair<int, int> >{{0, 0}};
- })
.set_attr<FCompute>("FCompute<cpu>", RandomContrast)
-.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
-.add_argument("data", "NDArray-or-Symbol", "The input.")
.add_arguments(RandomEnhanceParam::__FIELDS__());
-NNVM_REGISTER_OP(_image_random_saturation)
+MXNET_REGISTER_IMAGE_RND_AUG_OP(_image_random_saturation)
.describe(R"code()code" ADD_FILELINE)
-.set_num_inputs(1)
-.set_num_outputs(1)
.set_attr_parser(ParamParser<RandomEnhanceParam>)
-.set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& attrs) {
- return std::vector<ResourceRequest>{ResourceRequest::kRandom};
-})
-.set_attr<nnvm::FInferShape>("FInferShape", ImageShape)
-.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
-.set_attr<nnvm::FInplaceOption>("FInplaceOption",
- [](const NodeAttrs& attrs){
- return std::vector<std::pair<int, int> >{{0, 0}};
- })
.set_attr<FCompute>("FCompute<cpu>", RandomSaturation)
-.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
-.add_argument("data", "NDArray-or-Symbol", "The input.")
.add_arguments(RandomEnhanceParam::__FIELDS__());
-NNVM_REGISTER_OP(_image_random_hue)
+
+MXNET_REGISTER_IMAGE_RND_AUG_OP(_image_random_hue)
.describe(R"code()code" ADD_FILELINE)
-.set_num_inputs(1)
-.set_num_outputs(1)
.set_attr_parser(ParamParser<RandomEnhanceParam>)
-.set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& attrs) {
- return std::vector<ResourceRequest>{ResourceRequest::kRandom};
-})
-.set_attr<nnvm::FInferShape>("FInferShape", ImageShape)
-.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
.set_attr<FCompute>("FCompute<cpu>", RandomHue)
-.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
-.add_argument("data", "NDArray-or-Symbol", "The input.")
.add_arguments(RandomEnhanceParam::__FIELDS__());
-NNVM_REGISTER_OP(_image_random_color_jitter)
+
+MXNET_REGISTER_IMAGE_RND_AUG_OP(_image_random_color_jitter)
.describe(R"code()code" ADD_FILELINE)
-.set_num_inputs(1)
-.set_num_outputs(1)
.set_attr_parser(ParamParser<RandomColorJitterParam>)
-.set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& attrs) {
- return std::vector<ResourceRequest>{ResourceRequest::kRandom};
-})
-.set_attr<nnvm::FInferShape>("FInferShape", ImageShape)
-.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
.set_attr<FCompute>("FCompute<cpu>", RandomColorJitter)
-.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
-.add_argument("data", "NDArray-or-Symbol", "The input.")
.add_arguments(RandomColorJitterParam::__FIELDS__());
-NNVM_REGISTER_OP(_image_adjust_lighting)
+
+MXNET_REGISTER_IMAGE_AUG_OP(_image_adjust_lighting)
.describe(R"code(Adjust the lighting level of the input. Follow the AlexNet style.)code" ADD_FILELINE)
-.set_num_inputs(1)
-.set_num_outputs(1)
.set_attr_parser(ParamParser<AdjustLightingParam>)
-.set_attr<nnvm::FInferShape>("FInferShape", ImageShape)
-.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
-.set_attr<nnvm::FInplaceOption>("FInplaceOption",
- [](const NodeAttrs& attrs){
- return std::vector<std::pair<int, int> >{{0, 0}};
- })
.set_attr<FCompute>("FCompute<cpu>", AdjustLighting)
-.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
-.add_argument("data", "NDArray-or-Symbol", "The input.")
.add_arguments(AdjustLightingParam::__FIELDS__());
-NNVM_REGISTER_OP(_image_random_lighting)
+MXNET_REGISTER_IMAGE_RND_AUG_OP(_image_random_lighting)
.describe(R"code(Randomly add PCA noise. Follow the AlexNet style.)code" ADD_FILELINE)
-.set_num_inputs(1)
-.set_num_outputs(1)
.set_attr_parser(ParamParser<RandomLightingParam>)
-.set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& attrs) {
- return std::vector<ResourceRequest>{ResourceRequest::kRandom};
-})
-.set_attr<nnvm::FInferShape>("FInferShape", ImageShape)
-.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
-.set_attr<nnvm::FInplaceOption>("FInplaceOption",
- [](const NodeAttrs& attrs){
- return std::vector<std::pair<int, int> >{{0, 0}};
- })
.set_attr<FCompute>("FCompute<cpu>", RandomLighting)
-.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
-.add_argument("data", "NDArray-or-Symbol", "The input.")
.add_arguments(RandomLightingParam::__FIELDS__());
} // namespace image
--
To stop receiving notification emails like this one, please contact
"commits@mxnet.apache.org" <co...@mxnet.apache.org>.
[incubator-mxnet] 02/19: [WIP]]Vision (#8649)
Posted by jx...@apache.org.
This is an automated email from the ASF dual-hosted git repository.
jxie pushed a commit to branch vision
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git
commit 54fa02844cf9d2ec81f1cd5b4671aa6cc159ad96
Author: Hu Shiwen <ya...@gmail.com>
AuthorDate: Tue Nov 14 21:11:53 2017 +0800
[WIP]]Vision (#8649)
* not use opencv
---
src/operator/image/image_random-inl.h | 93 ++++++++++++++++++++++++-----------
src/operator/image/image_random.cc | 3 ++
2 files changed, 68 insertions(+), 28 deletions(-)
diff --git a/src/operator/image/image_random-inl.h b/src/operator/image/image_random-inl.h
index 027d587..9506607 100644
--- a/src/operator/image/image_random-inl.h
+++ b/src/operator/image/image_random-inl.h
@@ -29,7 +29,7 @@
#include <mxnet/base.h>
#include <opencv2/opencv.hpp>
#include <opencv2/core/mat.hpp>
-#include "mxnet/op_attr_types.h"
+#include "../mxnet_op.h"
#include "image_common.h"
@@ -44,6 +44,7 @@ struct RandomBrightnessParam : public dmlc::Parameter<RandomBrightnessParam> {
}
};
+enum ImageRandomResource { kRandom };
template<typename xpu>
static void RandomBrightness(const nnvm::NodeAttrs &attrs,
@@ -51,44 +52,80 @@ static void RandomBrightness(const nnvm::NodeAttrs &attrs,
const std::vector<TBlob> &inputs,
const std::vector<OpReqType> &req,
const std::vector<TBlob> &outputs) {
+ using namespace mshadow;
auto input = inputs[0];
auto output = outputs[0];
- int hight = input.shape_[0];
- int weight = input.shape_[1];
- int channel = input.shape_[2];
-
- auto input_mat = mat_convert(input, hight, weight, channel);
- auto output_mat = mat_convert(output, hight, weight, channel);
- //input_mat.convertTo(output_mat, -1, 1/255.0, 0);
- std::default_random_engine generator;
+ int channel = input.shape_[0];
+ int hight = input.shape_[1];
+ int weight = input.shape_[2];
+ Stream<xpu> *s = ctx.get_stream<xpu>();
+ Random<xpu> *prnd = ctx.requested[kRandom].get_random<xpu, real_t>(s);
+
const RandomBrightnessParam ¶m = nnvm::get<RandomBrightnessParam>(attrs.parsed);
- float alpha_b = 1.0 + std::uniform_real_distribution<float>(-param.max_brightness, param.max_brightness)(generator);
- output_mat.convertTo(output_mat, -1, alpha_b, 0);
+ float alpha_b = 1.0 + std::uniform_real_distribution<float>(-param.max_brightness, param.max_brightness)(prnd->GetRndEngine());
+ MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
+ MXNET_ASSIGN_REQ_SWITCH(req[0], Req,{
+ mxnet_op::Kernel<mxnet_op::op_with_req<mshadow::op::mul, Req>, xpu>::Launch(
+ s, inputs[0].Size(), outputs[0].dptr<DType>(), inputs[0].dptr<DType>(), DType(alpha_b));
+ });
+ });
+
}
+/*! \brief mul_add operator */
+struct mul_add {
+ /*! \brief map a, b, c to result using defined operation */
+ template<typename DType>
+ MSHADOW_XINLINE static DType Map(DType a, DType b, DType c) {
+ return a * b + c;
+ }
+};
+
+
template<typename xpu>
static void RandomContrast(const nnvm::NodeAttrs &attrs,
- const OpContext &ctx,
- const std::vector<TBlob> &inputs,
- const std::vector<OpReqType> &req,
- const std::vector<TBlob> &outputs) {
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+ using namespace mshadow;
auto input = inputs[0];
auto output = outputs[0];
- int hight = input.shape_[0];
- int weight = input.shape_[1];
- int channel = input.shape_[2];
-
- auto input_mat = mat_convert(input, hight, weight, channel);
- auto output_mat = mat_convert(output, hight, weight, channel);
- //input_mat.convertTo(output_mat, -1, 1/255.0, 0);
- std::default_random_engine generator;
+ int channel = input.shape_[0];
+ int hight = input.shape_[1];
+ int weight = input.shape_[2];
+ Stream<xpu> *s = ctx.get_stream<xpu>();
+ Random<xpu> *prnd = ctx.requested[kRandom].get_random<xpu, real_t>(s);
+
+
const RandomBrightnessParam ¶m = nnvm::get<RandomBrightnessParam>(attrs.parsed);
- float alpha_c = 1.0 + std::uniform_real_distribution<float>(-param.max_brightness, param.max_brightness)(generator);
- cv::Mat temp_;
- cv::cvtColor(input_mat, temp_, CV_RGB2GRAY);
- float gray_mean = cv::mean(temp_)[0];
- input_mat.convertTo(output_mat, -1, alpha_c, (1 - alpha_c) * gray_mean);
+ float alpha_c = 1.0 + std::uniform_real_distribution<float>(-param.max_brightness, param.max_brightness)(prnd->GetRndEngine());
+
+ const float R2YF = 0.299f;
+ const float G2YF = 0.587f;
+ const float B2YF = 0.114f;
+ static const float coeffs0[] = { R2YF, G2YF, B2YF };
+
+ MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
+ auto input_3d = input.FlatTo3D<xpu, DType>(s);
+ DType sum = (DType)0.0;
+ for (int c = 0; c < channel; ++c) {
+ for (int h = 0; h < hight; ++h) {
+ for (int w = 0; w < weight; ++w) {
+ sum += input_3d[c][h][w] * coeffs0[c];
+ }
+ }
+ }
+ float gray_mean = sum / (float)(hight * weight);
+ float beta = (1 - alpha_c) * gray_mean;
+
+ MXNET_ASSIGN_REQ_SWITCH(req[0], Req, {
+ mxnet_op::Kernel<mxnet_op::op_with_req<mul_add, Req>, xpu>::Launch(
+ s, inputs[0].Size(), outputs[0].dptr<DType>(), inputs[0].dptr<DType>(), DType(alpha_c), DType(beta));
+ });
+
+ });
}
diff --git a/src/operator/image/image_random.cc b/src/operator/image/image_random.cc
index 3777e43..ade43d8 100644
--- a/src/operator/image/image_random.cc
+++ b/src/operator/image/image_random.cc
@@ -39,6 +39,9 @@ NNVM_REGISTER_OP(_image_random_brightness)
.set_num_inputs(1)
.set_num_outputs(1)
.set_attr_parser(ParamParser<RandomBrightnessParam>)
+.set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& attrs) {
+ return std::vector<ResourceRequest>{ResourceRequest::kRandom};
+})
.set_attr<nnvm::FInferShape>("FInferShape", ElemwiseShape<1, 1>)
.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
.set_attr<FCompute>("FCompute<cpu>", RandomBrightness<cpu>)
--
To stop receiving notification emails like this one, please contact
"commits@mxnet.apache.org" <co...@mxnet.apache.org>.
[incubator-mxnet] 10/19: add test script
Posted by jx...@apache.org.
This is an automated email from the ASF dual-hosted git repository.
jxie pushed a commit to branch vision
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git
commit 1873367e17f5e4f434bd4f458bd0e426770be183
Author: Xingjian Shi <xs...@ust.hk>
AuthorDate: Tue Nov 21 11:07:21 2017 -0800
add test script
---
test_new_image_loader.py | 34 ++++++++++++++++++++++++++++++++++
1 file changed, 34 insertions(+)
diff --git a/test_new_image_loader.py b/test_new_image_loader.py
new file mode 100644
index 0000000..296869e
--- /dev/null
+++ b/test_new_image_loader.py
@@ -0,0 +1,34 @@
+import os
+os.environ['MXNET_CPU_WORKER_NTHREADS'] = '1'
+os.environ['OMP_NUM_THREADS'] = '1'
+import time
+import numpy as np
+import multiprocessing as mp
+import mxnet as mx
+from mxnet import gluon as gl
+from mxnet.gluon.data.vision import transforms
+
+if __name__ == '__main__':
+ M = 24
+ BS = 100
+
+ dataset = gl.data.vision.ImageFolderDataset('../256_ObjectCategories')
+ transform = transforms.Compose([transforms.ToTensor(),
+ transforms.RandomBrightness(1.0),
+ transforms.RandomContrast(1.0),
+ transforms.RandomSaturation(1.0),
+ transforms.Normalize([0, 0, 0], [1, 1, 1])])
+ dataset = dataset.transform_first(lambda x: transform(mx.image.center_crop(x, (224, 224))[0]))
+ data_loader = gl.data.DataLoader(dataset, BS, shuffle=True, num_workers=M)
+
+ N = len(dataset)
+
+ iterator = iter(data_loader)
+
+ tic = time.time()
+
+ for data, label in iterator:
+ data.wait_to_read()
+ print(data.shape)
+
+ print(N/(time.time() - tic))
--
To stop receiving notification emails like this one, please contact
"commits@mxnet.apache.org" <co...@mxnet.apache.org>.
[incubator-mxnet] 18/19: add comments and sanity check (#8901)
Posted by jx...@apache.org.
This is an automated email from the ASF dual-hosted git repository.
jxie pushed a commit to branch vision
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git
commit fc940fb9b0489fdfc6dc084aa0cce251391fa5d9
Author: Eric Junyuan Xie <pi...@users.noreply.github.com>
AuthorDate: Thu Nov 30 11:37:24 2017 -0800
add comments and sanity check (#8901)
---
python/mxnet/gluon/data/dataset.py | 56 ++++++-
python/mxnet/gluon/data/vision/transforms.py | 229 +++++++++++++++++++++------
tests/python/unittest/test_gluon_data.py | 21 +++
3 files changed, 251 insertions(+), 55 deletions(-)
diff --git a/python/mxnet/gluon/data/dataset.py b/python/mxnet/gluon/data/dataset.py
index f7ab395..4b97e43 100644
--- a/python/mxnet/gluon/data/dataset.py
+++ b/python/mxnet/gluon/data/dataset.py
@@ -41,12 +41,53 @@ class Dataset(object):
raise NotImplementedError
def transform(self, fn, lazy=True):
+ """Returns a new dataset with each sample transformed by the
+ transformer function `fn`.
+
+ Parameters
+ ----------
+ fn : callable
+ A transformer function that takes a sample as input and
+ returns the transformed sample.
+ lazy : bool, default True
+ If False, transforms all samples at once. Otherwise,
+ transforms each sample on demand. Note that if `fn`
+ is stochastic, you must set lazy to True or you will
+ get the same result on all epochs.
+
+ Returns
+ -------
+ Dataset
+ The transformed dataset.
+ """
trans = _LazyTransformDataset(self, fn)
if lazy:
return trans
return SimpleDataset([i for i in trans])
def transform_first(self, fn, lazy=True):
+ """Returns a new dataset with the first element of each sample
+ transformed by the transformer function `fn`.
+
+ This is useful, for example, when you only want to transform data
+ while keeping label as is.
+
+ Parameters
+ ----------
+ fn : callable
+ A transformer function that takes the first elemtn of a sample
+ as input and returns the transformed element.
+ lazy : bool, default True
+ If False, transforms all samples at once. Otherwise,
+ transforms each sample on demand. Note that if `fn`
+ is stochastic, you must set lazy to True or you will
+ get the same result on all epochs.
+
+ Returns
+ -------
+ Dataset
+ The transformed dataset.
+ """
def base_fn(x, *args):
if args:
return (fn(x),) + args
@@ -55,6 +96,13 @@ class Dataset(object):
class SimpleDataset(Dataset):
+ """Simple Dataset wrapper for lists and arrays.
+
+ Parameters
+ ----------
+ data : dataset-like object
+ Any object that implements `len()` and `[]`.
+ """
def __init__(self, data):
self._data = data
@@ -66,6 +114,7 @@ class SimpleDataset(Dataset):
class _LazyTransformDataset(Dataset):
+ """Lazily transformed dataset."""
def __init__(self, data, fn):
self._data = data
self._fn = fn
@@ -81,13 +130,14 @@ class _LazyTransformDataset(Dataset):
class ArrayDataset(Dataset):
- """A dataset of multiple arrays.
+ """A dataset that combines multiple dataset-like objects, e.g.
+ Datasets, lists, arrays, etc.
- The i-th sample is `(x1[i], x2[i], ...)`.
+ The i-th sample is defined as `(x1[i], x2[i], ...)`.
Parameters
----------
- *args : one or more arrays
+ *args : one or more dataset-like objects
The data arrays.
"""
def __init__(self, *args):
diff --git a/python/mxnet/gluon/data/vision/transforms.py b/python/mxnet/gluon/data/vision/transforms.py
index 931d644..8daf88e 100644
--- a/python/mxnet/gluon/data/vision/transforms.py
+++ b/python/mxnet/gluon/data/vision/transforms.py
@@ -20,11 +20,18 @@
from .. import dataset
from ...block import Block, HybridBlock
from ...nn import Sequential, HybridSequential
-from .... import ndarray, initializer
-from ....base import _Null
+from .... import ndarray, initializer, image
+from ....base import _Null, numeric_types
class Compose(Sequential):
+ """Sequentially composes multiple transforms.
+
+ Parameters
+ ----------
+ transforms : list of transform Blocks.
+ The list of transforms to be composed.
+ """
def __init__(self, transforms):
super(Compose, self).__init__()
transforms.append(None)
@@ -34,18 +41,25 @@ class Compose(Sequential):
hybrid.append(i)
continue
elif len(hybrid) == 1:
- self.register_child(hybrid[0])
+ self.add(hybrid[0])
elif len(hybrid) > 1:
hblock = HybridSequential()
for j in hybrid:
hblock.add(j)
- self.register_child(hblock)
+ self.add(hblock)
if i is not None:
- self.register_child(i)
+ self.add(i)
self.hybridize()
class Cast(HybridBlock):
+ """Cast input to a specific data type
+
+ Parameters
+ ----------
+ dtype : str, default 'float32'
+ The target data type, in string or `numpy.dtype`.
+ """
def __init__(self, dtype='float32'):
super(Cast, self).__init__()
self._dtype = dtype
@@ -55,6 +69,12 @@ class Cast(HybridBlock):
class ToTensor(HybridBlock):
+ """Converts an image NDArray to a tensor NDArray.
+
+ Converts an image NDArray of shape (H x W x C) in the range
+ [0, 255] to a float32 tensor NDArray of shape (C x H x W) in
+ the range [0, 1).
+ """
def __init__(self):
super(ToTensor, self).__init__()
@@ -63,6 +83,23 @@ class ToTensor(HybridBlock):
class Normalize(HybridBlock):
+ """Normalize an tensor of shape (C x H x W) with mean and
+ standard deviation.
+
+ Given mean `(m1, ..., mn)` and std `(s1, ..., sn)` for `n` channels,
+ this transform normalizes each channel of the input tensor with::
+
+ output[i] = (input[i] - mi) / si
+
+ If mean or std is scalar, the same value will be applied to all channels.
+
+ Parameters
+ ----------
+ mean : float or tuple of floats
+ The mean values.
+ std : float or tuple of floats
+ The standard deviation values.
+ """
def __init__(self, mean, std):
super(Normalize, self).__init__()
self._mean = mean
@@ -72,101 +109,189 @@ class Normalize(HybridBlock):
return F.image.normalize(x, self._mean, self._std)
-class RandomResizedCrop(HybridBlock):
- def __init__(self, size, area=(0.08, 1.0), ratio=(3.0/4.0, 4.0/3.0),
+class RandomResizedCrop(Block):
+ """Crop the input image with random scale and aspect ratio.
+
+ Makes a crop of the original image with random size (default: 0.08
+ to 1.0 of the original image size) and random aspect ratio (default:
+ 3/4 to 4/3), then resize it to the specified size.
+
+ Parameters
+ ----------
+ size : int or tuple of (W, H)
+ Size of the final output.
+ scale : tuple of two floats
+ If scale is `(min_area, max_area)`, the cropped image's area will
+ range from min_area to max_area of the original image's area
+ ratio : tuple of two floats
+ Range of aspect ratio of the cropped image before resizing.
+ interpolation : int
+ Interpolation method for resizing. By default uses bilinear
+ interpolation. See OpenCV's resize function for available choices.
+ """
+ def __init__(self, size, scale=(0.08, 1.0), ratio=(3.0/4.0, 4.0/3.0),
interpolation=2):
super(RandomResizedCrop, self).__init__()
- self._args = (size, area, ratio, interpolation)
-
- def hybrid_forward(self, F, x):
- return F.image.random_resized_crop(x, *self._args)
-
-
-class CenterCrop(HybridBlock):
- def __init__(self, size):
+ if isinstance(size, numeric_types):
+ size = (size, size)
+ self._args = (size, scale[0], ratio, interpolation)
+
+ def forward(self, x):
+ return image.random_size_crop(x, *self._args)[0]
+
+
+class CenterCrop(Block):
+ """Crops the image `src` to the given `size` by trimming on all four
+ sides and preserving the center of the image. Upsamples if `src` is
+ smaller than `size`.
+
+ Parameters
+ ----------
+ size : int or tuple of (W, H)
+ Size of output image.
+ interpolation : int
+ Interpolation method for resizing. By default uses bilinear
+ interpolation. See OpenCV's resize function for available choices.
+ """
+ def __init__(self, size, interpolation=2):
super(CenterCrop, self).__init__()
- self._size = size
+ if isinstance(size, numeric_types):
+ size = (size, size)
+ self._args = (size, interpolation)
+
+ def forward(self, x):
+ return image.center_crop(x, *self._args)[0]
- def hybrid_forward(self, F, x):
- return F.image.center_crop(x, size)
+class Resize(Block):
+ """Resize an image to the given size.
-class Resize(HybridBlock):
+ Parameters
+ ----------
+ size : int or tuple of (W, H)
+ Size of output image.
+ interpolation : int
+ Interpolation method for resizing. By default uses bilinear
+ interpolation. See OpenCV's resize function for available choices.
+ """
def __init__(self, size, interpolation=2):
super(Resize, self).__init__()
- self._args = (size, interpolation)
+ if isinstance(size, numeric_types):
+ size = (size, size)
+ self._args = tuple(size) + (interpolation,)
+
+ def forward(self, x):
+ return image.imresize(x, *self._args)
+
+
+class RandomHorizontalFlip(HybridBlock):
+ """Randomly flip the input image horizontally with a probability
+ of 0.5.
+ """
+ def __init__(self):
+ super(RandomHorizontalFlip, self).__init__()
def hybrid_forward(self, F, x):
- return F.image.resize(x, *self._args)
+ return F.image.random_horizontal_flip(x)
-class RandomFlip(HybridBlock):
- def __init__(self, axis=1):
- super(RandomFlip, self).__init__()
- self._axis = axis
+class RandomVerticalFlip(HybridBlock):
+ """Randomly flip the input image vertically with a probability
+ of 0.5.
+ """
+ def __init__(self):
+ super(RandomVerticalFlip, self).__init__()
def hybrid_forward(self, F, x):
- return F.image.random_flip(x, self._axis)
+ return F.image.random_vertical_flip(x)
class RandomBrightness(HybridBlock):
- def __init__(self, max_brightness):
+ """Randomly jitters image brightness with a factor
+ chosen from `[max(0, 1 - brightness), 1 + brightness]`.
+ """
+ def __init__(self, brightness):
super(RandomBrightness, self).__init__()
- self._max_brightness = max_brightness
+ self._args = (max(0, 1-brightness), 1+brightness)
def hybrid_forward(self, F, x):
- return F.image.random_brightness(x, self._max_brightness)
+ return F.image.random_brightness(x, *self._args)
class RandomContrast(HybridBlock):
- def __init__(self, max_contrast):
+ """Randomly jitters image contrast with a factor
+ chosen from `[max(0, 1 - contrast), 1 + contrast]`.
+ """
+ def __init__(self, contrast):
super(RandomContrast, self).__init__()
- self._max_contrast = max_contrast
+ self._args = (max(0, 1-contrast), 1+contrast)
def hybrid_forward(self, F, x):
- return F.image.random_contrast(x, self._max_contrast)
+ return F.image.random_contrast(x, *self._args)
class RandomSaturation(HybridBlock):
- def __init__(self, max_saturation):
+ """Randomly jitters image saturation with a factor
+ chosen from `[max(0, 1 - saturation), 1 + saturation]`.
+ """
+ def __init__(self, saturation):
super(RandomSaturation, self).__init__()
- self._max_saturation = max_saturation
+ self._args = (max(0, 1-saturation), 1+saturation)
def hybrid_forward(self, F, x):
- return F.image.random_saturation(x, self._max_saturation)
+ return F.image.random_saturation(x, *self._args)
class RandomHue(HybridBlock):
- def __init__(self, max_hue):
+ """Randomly jitters image hue with a factor
+ chosen from `[max(0, 1 - hue), 1 + hue]`.
+ """
+ def __init__(self, hue):
super(RandomHue, self).__init__()
- self._max_hue = max_hue
+ self._args = (max(0, 1-hue), 1+hue)
def hybrid_forward(self, F, x):
- return F.image.random_hue(x, self._max_hue)
+ return F.image.random_hue(x, *self._args)
class RandomColorJitter(HybridBlock):
- def __init__(self, max_brightness=0, max_contrast=0, max_saturation=0, max_hue=0):
+ """Randomly jitters the brightness, contrast, saturation, and hue
+ of an image.
+
+ Parameters
+ ----------
+ brightness : float
+ How much to jitter brightness. brightness factor is randomly
+ chosen from `[max(0, 1 - brightness), 1 + brightness]`.
+ contrast : float
+ How much to jitter contrast. contrast factor is randomly
+ chosen from `[max(0, 1 - contrast), 1 + contrast]`.
+ saturation : float
+ How much to jitter saturation. saturation factor is randomly
+ chosen from `[max(0, 1 - saturation), 1 + saturation]`.
+ hue : float
+ How much to jitter hue. hue factor is randomly
+ chosen from `[max(0, 1 - hue), 1 + hue]`.
+ """
+ def __init__(self, brightness=0, contrast=0, saturation=0, hue=0):
super(RandomColorJitter, self).__init__()
- self._args = (max_brightness, max_contrast, max_saturation, max_hue)
+ self._args = (brightness, contrast, saturation, hue)
def hybrid_forward(self, F, x):
return F.image.random_color_jitter(x, *self._args)
-class AdjustLighting(HybridBlock):
- def __init__(self, alpha_rgb=_Null, eigval=_Null, eigvec=_Null):
- super(AdjustLighting, self).__init__()
- self._args = (alpha_rgb, eigval, eigvec)
-
- def hybrid_forward(self, F, x):
- return F.image.adjust_lighting(x, *self._args)
-
-
class RandomLighting(HybridBlock):
- def __init__(self, alpha_std=_Null, eigval=_Null, eigvec=_Null):
+ """Add AlexNet-style PCA-based noise to an image.
+
+ Parameters
+ ----------
+ alpha : float
+ Intensity of the image.
+ """
+ def __init__(self, alpha):
super(RandomLighting, self).__init__()
- self._args = (alpha_std, eigval, eigvec)
+ self._alpha = alpha
def hybrid_forward(self, F, x):
- return F.image.random_lighting(x, *self._args)
\ No newline at end of file
+ return F.image.random_lighting(x, self._alpha)
diff --git a/tests/python/unittest/test_gluon_data.py b/tests/python/unittest/test_gluon_data.py
index 63c5d28..c72ef7c 100644
--- a/tests/python/unittest/test_gluon_data.py
+++ b/tests/python/unittest/test_gluon_data.py
@@ -107,6 +107,27 @@ def test_multi_worker():
assert (batch.asnumpy() == i).all()
+def test_transformer():
+ from mxnet.gluon.data.vision import transforms
+
+ transform = transforms.Compose([
+ transforms.Resize(300),
+ transforms.CenterCrop(256),
+ transforms.RandomResizedCrop(224),
+ transforms.RandomHorizontalFlip(),
+ transforms.RandomColorJitter(0.1, 0.1, 0.1, 0.1),
+ transforms.RandomBrightness(0.1),
+ transforms.RandomContrast(0.1),
+ transforms.RandomSaturation(0.1),
+ transforms.RandomHue(0.1),
+ transforms.RandomLighting(0.1),
+ transforms.ToTensor(),
+ transforms.Normalize([0, 0, 0], [1, 1, 1])])
+
+ transform(mx.nd.ones((245, 480, 3), dtype='uint8')).wait_to_read()
+
+
if __name__ == '__main__':
+ test_transformer()
import nose
nose.runmodule()
--
To stop receiving notification emails like this one, please contact
"commits@mxnet.apache.org" <co...@mxnet.apache.org>.
[incubator-mxnet] 13/19: [Image] add random lighting (#8779)
Posted by jx...@apache.org.
This is an automated email from the ASF dual-hosted git repository.
jxie pushed a commit to branch vision
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git
commit 3c99773c9a58be7f64147d257f14357f5856e5d7
Author: Xingjian Shi <xs...@ust.hk>
AuthorDate: Wed Nov 22 15:53:37 2017 -0800
[Image] add random lighting (#8779)
* add random lighting
* fix
---
python/mxnet/gluon/data/vision/transforms.py | 19 +++++
src/operator/image/image_random-inl.h | 95 +++++++++++++++++++++++++
src/operator/image/image_random.cc | 43 +++++++++--
tests/python/unittest/test_gluon_data_vision.py | 40 +++++++++++
4 files changed, 191 insertions(+), 6 deletions(-)
diff --git a/python/mxnet/gluon/data/vision/transforms.py b/python/mxnet/gluon/data/vision/transforms.py
index e1deef6..931d644 100644
--- a/python/mxnet/gluon/data/vision/transforms.py
+++ b/python/mxnet/gluon/data/vision/transforms.py
@@ -21,6 +21,7 @@ from .. import dataset
from ...block import Block, HybridBlock
from ...nn import Sequential, HybridSequential
from .... import ndarray, initializer
+from ....base import _Null
class Compose(Sequential):
@@ -151,3 +152,21 @@ class RandomColorJitter(HybridBlock):
def hybrid_forward(self, F, x):
return F.image.random_color_jitter(x, *self._args)
+
+
+class AdjustLighting(HybridBlock):
+ def __init__(self, alpha_rgb=_Null, eigval=_Null, eigvec=_Null):
+ super(AdjustLighting, self).__init__()
+ self._args = (alpha_rgb, eigval, eigvec)
+
+ def hybrid_forward(self, F, x):
+ return F.image.adjust_lighting(x, *self._args)
+
+
+class RandomLighting(HybridBlock):
+ def __init__(self, alpha_std=_Null, eigval=_Null, eigvec=_Null):
+ super(RandomLighting, self).__init__()
+ self._args = (alpha_std, eigval, eigvec)
+
+ def hybrid_forward(self, F, x):
+ return F.image.random_lighting(x, *self._args)
\ No newline at end of file
diff --git a/src/operator/image/image_random-inl.h b/src/operator/image/image_random-inl.h
index f823c8c..ebbf60a 100644
--- a/src/operator/image/image_random-inl.h
+++ b/src/operator/image/image_random-inl.h
@@ -26,6 +26,7 @@
#define MXNET_OPERATOR_IMAGE_IMAGE_RANDOM_INL_H_
#include <mxnet/base.h>
+#include <algorithm>
#include <vector>
#include <opencv2/opencv.hpp>
#include <opencv2/core/mat.hpp>
@@ -290,11 +291,105 @@ static void RandomColorJitter(const nnvm::NodeAttrs &attrs,
const std::vector<TBlob> &outputs) {
}
+struct AdjustLightingParam : public dmlc::Parameter<AdjustLightingParam> {
+ nnvm::Tuple<float> alpha_rgb;
+ nnvm::Tuple<float> eigval;
+ nnvm::Tuple<float> eigvec;
+ DMLC_DECLARE_PARAMETER(AdjustLightingParam) {
+ DMLC_DECLARE_FIELD(alpha_rgb)
+ .set_default({0, 0, 0})
+ .describe("The lighting alphas for the R, G, B channels.");
+ DMLC_DECLARE_FIELD(eigval)
+ .describe("Eigen value.")
+ .set_default({ 55.46, 4.794, 1.148 });
+ DMLC_DECLARE_FIELD(eigvec)
+ .describe("Eigen vector.")
+ .set_default({ -0.5675, 0.7192, 0.4009,
+ -0.5808, -0.0045, -0.8140,
+ -0.5808, -0.0045, -0.8140 });
+ }
+};
+
+struct RandomLightingParam : public dmlc::Parameter<RandomLightingParam> {
+ float alpha_std;
+ nnvm::Tuple<float> eigval;
+ nnvm::Tuple<float> eigvec;
+ DMLC_DECLARE_PARAMETER(RandomLightingParam) {
+ DMLC_DECLARE_FIELD(alpha_std)
+ .set_default(0.05)
+ .describe("Level of the lighting noise.");
+ DMLC_DECLARE_FIELD(eigval)
+ .describe("Eigen value.")
+ .set_default({ 55.46, 4.794, 1.148 });
+ DMLC_DECLARE_FIELD(eigvec)
+ .describe("Eigen vector.")
+ .set_default({ -0.5675, 0.7192, 0.4009,
+ -0.5808, -0.0045, -0.8140,
+ -0.5808, -0.0045, -0.8140 });
+ }
+};
+
+void AdjustLightingImpl(uint8_t* dst, const uint8_t* src,
+ float alpha_r, float alpha_g, float alpha_b,
+ const nnvm::Tuple<float> eigval, const nnvm::Tuple<float> eigvec,
+ int H, int W) {
+ alpha_r *= eigval[0];
+ alpha_g *= eigval[1];
+ alpha_b *= eigval[2];
+ float pca_r = alpha_r * eigvec[0] + alpha_g * eigvec[1] + alpha_b * eigvec[2];
+ float pca_g = alpha_r * eigvec[3] + alpha_g * eigvec[4] + alpha_b * eigvec[5];
+ float pca_b = alpha_r * eigvec[6] + alpha_g * eigvec[7] + alpha_b * eigvec[8];
+ for (int i = 0; i < H * W; i++) {
+ int base_ind = 3 * i;
+ float in_r = static_cast<float>(src[base_ind]);
+ float in_g = static_cast<float>(src[base_ind + 1]);
+ float in_b = static_cast<float>(src[base_ind + 2]);
+ dst[base_ind] = std::min(255, std::max(0, static_cast<int>(in_r + pca_r)));
+ dst[base_ind + 1] = std::min(255, std::max(0, static_cast<int>(in_g + pca_g)));
+ dst[base_ind + 2] = std::min(255, std::max(0, static_cast<int>(in_b + pca_b)));
+ }
+}
+
+static void AdjustLighting(const nnvm::NodeAttrs &attrs,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+ using namespace mshadow;
+ const AdjustLightingParam ¶m = nnvm::get<AdjustLightingParam>(attrs.parsed);
+ CHECK_EQ(param.eigval.ndim(), 3) << "There should be 3 numbers in the eigval.";
+ CHECK_EQ(param.eigvec.ndim(), 9) << "There should be 9 numbers in the eigvec.";
+ CHECK_EQ(inputs[0].ndim(), 3);
+ CHECK_EQ(inputs[0].size(2), 3);
+ int H = inputs[0].size(0);
+ int W = inputs[0].size(1);
+ AdjustLightingImpl(outputs[0].dptr<uint8_t>(), inputs[0].dptr<uint8_t>(),
+ param.alpha_rgb[0], param.alpha_rgb[1], param.alpha_rgb[2],
+ param.eigval, param.eigvec, H, W);
+}
+
static void RandomLighting(const nnvm::NodeAttrs &attrs,
const OpContext &ctx,
const std::vector<TBlob> &inputs,
const std::vector<OpReqType> &req,
const std::vector<TBlob> &outputs) {
+ using namespace mshadow;
+ const RandomLightingParam ¶m = nnvm::get<RandomLightingParam>(attrs.parsed);
+ CHECK_EQ(param.eigval.ndim(), 3) << "There should be 3 numbers in the eigval.";
+ CHECK_EQ(param.eigvec.ndim(), 9) << "There should be 9 numbers in the eigvec.";
+ CHECK_EQ(inputs[0].ndim(), 3);
+ CHECK_EQ(inputs[0].size(2), 3);
+ int H = inputs[0].size(0);
+ int W = inputs[0].size(1);
+ Stream<cpu> *s = ctx.get_stream<cpu>();
+ Random<cpu> *prnd = ctx.requested[0].get_random<cpu, real_t>(s);
+ std::normal_distribution<float> dist(0, param.alpha_std);
+ float alpha_r = dist(prnd->GetRndEngine());
+ float alpha_g = dist(prnd->GetRndEngine());
+ float alpha_b = dist(prnd->GetRndEngine());
+ AdjustLightingImpl(outputs[0].dptr<uint8_t>(), inputs[0].dptr<uint8_t>(),
+ alpha_r, alpha_g, alpha_b,
+ param.eigval, param.eigvec, H, W);
}
diff --git a/src/operator/image/image_random.cc b/src/operator/image/image_random.cc
index 7ff7328..5b47f50 100644
--- a/src/operator/image/image_random.cc
+++ b/src/operator/image/image_random.cc
@@ -35,9 +35,6 @@ NNVM_REGISTER_OP(_image_to_tensor)
.describe(R"code()code" ADD_FILELINE)
.set_num_inputs(1)
.set_num_outputs(1)
-.set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& attrs) {
- return std::vector<ResourceRequest>{ResourceRequest::kRandom};
-})
.set_attr<nnvm::FInferShape>("FInferShape", ToTensorShape)
.set_attr<nnvm::FInferType>("FInferType", ToTensorType)
.set_attr<FCompute>("FCompute<cpu>", ToTensor)
@@ -51,9 +48,6 @@ NNVM_REGISTER_OP(_image_normalize)
.set_num_inputs(1)
.set_num_outputs(1)
.set_attr_parser(ParamParser<NormalizeParam>)
-.set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& attrs) {
- return std::vector<ResourceRequest>{ResourceRequest::kRandom};
-})
.set_attr<nnvm::FInferShape>("FInferShape", ElemwiseShape<1, 1>)
.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
.set_attr<nnvm::FInplaceOption>("FInplaceOption",
@@ -126,5 +120,42 @@ NNVM_REGISTER_OP(_image_random_saturation)
.add_argument("data", "NDArray-or-Symbol", "The input.")
.add_arguments(RandomSaturationParam::__FIELDS__());
+DMLC_REGISTER_PARAMETER(AdjustLightingParam);
+NNVM_REGISTER_OP(_image_adjust_lighting)
+.describe(R"code(Adjust the lighting level of the input. Follow the AlexNet style.)code" ADD_FILELINE)
+.set_num_inputs(1)
+.set_num_outputs(1)
+.set_attr_parser(ParamParser<AdjustLightingParam>)
+.set_attr<nnvm::FInferShape>("FInferShape", ElemwiseShape<1, 1>)
+.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
+.set_attr<nnvm::FInplaceOption>("FInplaceOption",
+ [](const NodeAttrs& attrs){
+ return std::vector<std::pair<int, int> >{{0, 0}};
+ })
+.set_attr<FCompute>("FCompute<cpu>", AdjustLighting)
+.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
+.add_argument("data", "NDArray-or-Symbol", "The input.")
+.add_arguments(AdjustLightingParam::__FIELDS__());
+
+DMLC_REGISTER_PARAMETER(RandomLightingParam);
+NNVM_REGISTER_OP(_image_random_lighting)
+.describe(R"code(Randomly add PCA noise. Follow the AlexNet style.)code" ADD_FILELINE)
+.set_num_inputs(1)
+.set_num_outputs(1)
+.set_attr_parser(ParamParser<RandomLightingParam>)
+.set_attr<FResourceRequest>("FResourceRequest", [](const NodeAttrs& attrs) {
+ return std::vector<ResourceRequest>{ResourceRequest::kRandom};
+})
+.set_attr<nnvm::FInferShape>("FInferShape", ElemwiseShape<1, 1>)
+.set_attr<nnvm::FInferType>("FInferType", ElemwiseType<1, 1>)
+.set_attr<nnvm::FInplaceOption>("FInplaceOption",
+ [](const NodeAttrs& attrs){
+ return std::vector<std::pair<int, int> >{{0, 0}};
+ })
+.set_attr<FCompute>("FCompute<cpu>", RandomLighting)
+.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
+.add_argument("data", "NDArray-or-Symbol", "The input.")
+.add_arguments(RandomLightingParam::__FIELDS__());
+
} // namespace op
} // namespace mxnet
diff --git a/tests/python/unittest/test_gluon_data_vision.py b/tests/python/unittest/test_gluon_data_vision.py
new file mode 100644
index 0000000..0c9e5c1
--- /dev/null
+++ b/tests/python/unittest/test_gluon_data_vision.py
@@ -0,0 +1,40 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements. See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership. The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License. You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied. See the License for the
+# specific language governing permissions and limitations
+# under the License.
+from __future__ import print_function
+import mxnet as mx
+import mxnet.ndarray as nd
+import numpy as np
+from mxnet import gluon
+from mxnet.gluon.data.vision.transforms import AdjustLighting
+from mxnet.test_utils import assert_almost_equal
+
+def test_adjust_lighting():
+ data_in = np.random.uniform(0, 255, (300, 300, 3)).astype(dtype=np.uint8)
+ alpha_rgb = [0.05, 0.06, 0.07]
+ eigval = np.array([55.46, 4.794, 1.148])
+ eigvec = np.array([[-0.5675, 0.7192, 0.4009],
+ [-0.5808, -0.0045, -0.8140],
+ [-0.5808, -0.0045, -0.8140]])
+ f = AdjustLighting(alpha_rgb=alpha_rgb, eigval=eigval.ravel().tolist(), eigvec=eigvec.ravel().tolist())
+ out_nd = f(nd.array(data_in, dtype=np.uint8))
+ out_gt = np.clip(data_in.astype(np.float32)
+ + np.dot(eigvec * alpha_rgb, eigval.reshape((3, 1))).reshape((1, 1, 3)), 0, 255).astype(np.uint8)
+ assert_almost_equal(out_nd.asnumpy(), out_gt)
+
+if __name__ == '__main__':
+ import nose
+ nose.runmodule()
--
To stop receiving notification emails like this one, please contact
"commits@mxnet.apache.org" <co...@mxnet.apache.org>.
[incubator-mxnet] 19/19: [Vision] add test cases for flip, normalize,
to_tensor (#8919)
Posted by jx...@apache.org.
This is an automated email from the ASF dual-hosted git repository.
jxie pushed a commit to branch vision
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git
commit 8eec5369b3b8a7ff0cff9a8e9a0cdc1cd110734f
Author: Yizhi Liu <ja...@gmail.com>
AuthorDate: Tue Dec 12 13:28:41 2017 -0800
[Vision] add test cases for flip, normalize, to_tensor (#8919)
* [vision] ut for to_tensor, normalize, flip
* [vision] fix flip
* [vision] flip name
* [vision] test non-random flip op
* remove transform.FlipXXXX
---
python/mxnet/gluon/data/vision/transforms.py | 16 +++---
src/operator/image/image_random-inl.h | 68 +++++++++++++++++--------
src/operator/image/image_random.cc | 16 ++++--
tests/python/unittest/test_gluon_data_vision.py | 41 ++++++++++-----
4 files changed, 96 insertions(+), 45 deletions(-)
diff --git a/python/mxnet/gluon/data/vision/transforms.py b/python/mxnet/gluon/data/vision/transforms.py
index 8daf88e..38eb690 100644
--- a/python/mxnet/gluon/data/vision/transforms.py
+++ b/python/mxnet/gluon/data/vision/transforms.py
@@ -184,26 +184,26 @@ class Resize(Block):
return image.imresize(x, *self._args)
-class RandomHorizontalFlip(HybridBlock):
- """Randomly flip the input image horizontally with a probability
+class RandomFlipLeftRight(HybridBlock):
+ """Randomly flip the input image left to right with a probability
of 0.5.
"""
def __init__(self):
- super(RandomHorizontalFlip, self).__init__()
+ super(RandomFlipLeftRight, self).__init__()
def hybrid_forward(self, F, x):
- return F.image.random_horizontal_flip(x)
+ return F.image.random_flip_left_right(x)
-class RandomVerticalFlip(HybridBlock):
- """Randomly flip the input image vertically with a probability
+class RandomFlipTopBottom(HybridBlock):
+ """Randomly flip the input image top to bottom with a probability
of 0.5.
"""
def __init__(self):
- super(RandomVerticalFlip, self).__init__()
+ super(RandomFlipTopBottom, self).__init__()
def hybrid_forward(self, F, x):
- return F.image.random_vertical_flip(x)
+ return F.image.random_flip_top_bottom(x)
class RandomBrightness(HybridBlock):
diff --git a/src/operator/image/image_random-inl.h b/src/operator/image/image_random-inl.h
index cbc7f40..ec96149 100644
--- a/src/operator/image/image_random-inl.h
+++ b/src/operator/image/image_random-inl.h
@@ -85,14 +85,6 @@ void ToTensor(const nnvm::NodeAttrs &attrs,
});
}
-inline bool TensorShape(const nnvm::NodeAttrs& attrs,
- std::vector<TShape> *in_attrs,
- std::vector<TShape> *out_attrs) {
- TShape& dshape = (*in_attrs)[0];
- SHAPE_ASSIGN_CHECK(*out_attrs, 0, dshape);
- return true;
-}
-
struct NormalizeParam : public dmlc::Parameter<NormalizeParam> {
nnvm::Tuple<float> mean;
nnvm::Tuple<float> std;
@@ -179,16 +171,16 @@ inline bool ImageShape(const nnvm::NodeAttrs& attrs,
return true;
}
-template<typename DType>
-void FlipImpl(const TShape &shape, DType *src, DType *dst, int axis) {
+template<typename DType, int axis>
+void FlipImpl(const TShape &shape, DType *src, DType *dst) {
int head = 1, mid = shape[axis], tail = 1;
for (int i = 0; i < axis; ++i) head *= shape[i];
for (int i = axis+1; i < shape.ndim(); ++i) tail *= shape[i];
for (int i = 0; i < head; ++i) {
- for (int j = 0; j < (mid >>2); ++j) {
- int idx1 = (i*mid + j)*tail;
- int idx2 = idx1 + (mid - (j<<2))*tail;
+ for (int j = 0; j < (mid >> 1); ++j) {
+ int idx1 = (i*mid + j) * tail;
+ int idx2 = idx1 + (mid-(j << 1)-1) * tail;
for (int k = 0; k < tail; ++k, ++idx1, ++idx2) {
DType tmp = src[idx1];
dst[idx1] = src[idx2];
@@ -198,7 +190,31 @@ void FlipImpl(const TShape &shape, DType *src, DType *dst, int axis) {
}
}
-void RandomHorizontalFlip(
+void FlipLeftRight(const nnvm::NodeAttrs &attrs,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+ using namespace mshadow;
+ MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
+ FlipImpl<DType, 1>(inputs[0].shape_, inputs[0].dptr<DType>(),
+ outputs[0].dptr<DType>());
+ });
+}
+
+void FlipTopBottom(const nnvm::NodeAttrs &attrs,
+ const OpContext &ctx,
+ const std::vector<TBlob> &inputs,
+ const std::vector<OpReqType> &req,
+ const std::vector<TBlob> &outputs) {
+ using namespace mshadow;
+ MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
+ FlipImpl<DType, 0>(inputs[0].shape_, inputs[0].dptr<DType>(),
+ outputs[0].dptr<DType>());
+ });
+}
+
+void RandomFlipLeftRight(
const nnvm::NodeAttrs &attrs,
const OpContext &ctx,
const std::vector<TBlob> &inputs,
@@ -207,14 +223,19 @@ void RandomHorizontalFlip(
using namespace mshadow;
Stream<cpu> *s = ctx.get_stream<cpu>();
Random<cpu> *prnd = ctx.requested[0].get_random<cpu, float>(s);
- if (std::bernoulli_distribution()(prnd->GetRndEngine())) return;
MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
- FlipImpl(inputs[0].shape_, inputs[0].dptr<DType>(),
- outputs[0].dptr<DType>(), 1);
+ if (std::bernoulli_distribution()(prnd->GetRndEngine())) {
+ if (outputs[0].dptr_ != inputs[0].dptr_) {
+ std::memcpy(outputs[0].dptr_, inputs[0].dptr_, inputs[0].Size() * sizeof(DType));
+ }
+ } else {
+ FlipImpl<DType, 1>(inputs[0].shape_, inputs[0].dptr<DType>(),
+ outputs[0].dptr<DType>());
+ }
});
}
-void RandomVerticalFlip(
+void RandomFlipTopBottom(
const nnvm::NodeAttrs &attrs,
const OpContext &ctx,
const std::vector<TBlob> &inputs,
@@ -223,10 +244,15 @@ void RandomVerticalFlip(
using namespace mshadow;
Stream<cpu> *s = ctx.get_stream<cpu>();
Random<cpu> *prnd = ctx.requested[0].get_random<cpu, float>(s);
- if (std::bernoulli_distribution()(prnd->GetRndEngine())) return;
MSHADOW_TYPE_SWITCH(outputs[0].type_flag_, DType, {
- FlipImpl(inputs[0].shape_, inputs[0].dptr<DType>(),
- outputs[0].dptr<DType>(), 0);
+ if (std::bernoulli_distribution()(prnd->GetRndEngine())) {
+ if (outputs[0].dptr_ != inputs[0].dptr_) {
+ std::memcpy(outputs[0].dptr_, inputs[0].dptr_, inputs[0].Size() * sizeof(DType));
+ }
+ } else {
+ FlipImpl<DType, 0>(inputs[0].shape_, inputs[0].dptr<DType>(),
+ outputs[0].dptr<DType>());
+ }
});
}
diff --git a/src/operator/image/image_random.cc b/src/operator/image/image_random.cc
index 481dfce..26f520b 100644
--- a/src/operator/image/image_random.cc
+++ b/src/operator/image/image_random.cc
@@ -48,7 +48,6 @@ NNVM_REGISTER_OP(_image_to_tensor)
.set_attr<nnvm::FGradient>("FGradient", ElemwiseGradUseNone{ "_copy" })
.add_argument("data", "NDArray-or-Symbol", "The input.");
-
NNVM_REGISTER_OP(_image_normalize)
.describe(R"code()code" ADD_FILELINE)
.set_num_inputs(1)
@@ -65,11 +64,21 @@ NNVM_REGISTER_OP(_image_normalize)
.add_argument("data", "NDArray-or-Symbol", "The input.")
.add_arguments(NormalizeParam::__FIELDS__());
+MXNET_REGISTER_IMAGE_AUG_OP(_image_flip_left_right)
+.describe(R"code()code" ADD_FILELINE)
+.set_attr<FCompute>("FCompute<cpu>", FlipLeftRight);
+
+MXNET_REGISTER_IMAGE_RND_AUG_OP(_image_random_flip_left_right)
+.describe(R"code()code" ADD_FILELINE)
+.set_attr<FCompute>("FCompute<cpu>", RandomFlipLeftRight);
-MXNET_REGISTER_IMAGE_RND_AUG_OP(_image_random_horizontal_flip)
+MXNET_REGISTER_IMAGE_AUG_OP(_image_flip_top_bottom)
.describe(R"code()code" ADD_FILELINE)
-.set_attr<FCompute>("FCompute<cpu>", RandomHorizontalFlip);
+.set_attr<FCompute>("FCompute<cpu>", FlipTopBottom);
+MXNET_REGISTER_IMAGE_RND_AUG_OP(_image_random_flip_top_bottom)
+.describe(R"code()code" ADD_FILELINE)
+.set_attr<FCompute>("FCompute<cpu>", RandomFlipTopBottom);
MXNET_REGISTER_IMAGE_RND_AUG_OP(_image_random_brightness)
.describe(R"code()code" ADD_FILELINE)
@@ -77,7 +86,6 @@ MXNET_REGISTER_IMAGE_RND_AUG_OP(_image_random_brightness)
.set_attr<FCompute>("FCompute<cpu>", RandomBrightness)
.add_arguments(RandomEnhanceParam::__FIELDS__());
-
MXNET_REGISTER_IMAGE_RND_AUG_OP(_image_random_contrast)
.describe(R"code()code" ADD_FILELINE)
.set_attr_parser(ParamParser<RandomEnhanceParam>)
diff --git a/tests/python/unittest/test_gluon_data_vision.py b/tests/python/unittest/test_gluon_data_vision.py
index 0c9e5c1..5e9ff87 100644
--- a/tests/python/unittest/test_gluon_data_vision.py
+++ b/tests/python/unittest/test_gluon_data_vision.py
@@ -18,22 +18,39 @@ from __future__ import print_function
import mxnet as mx
import mxnet.ndarray as nd
import numpy as np
+from PIL import Image
from mxnet import gluon
-from mxnet.gluon.data.vision.transforms import AdjustLighting
+from mxnet.gluon.data.vision import transforms
from mxnet.test_utils import assert_almost_equal
+from mxnet.test_utils import almost_equal
-def test_adjust_lighting():
+def test_to_tensor():
data_in = np.random.uniform(0, 255, (300, 300, 3)).astype(dtype=np.uint8)
- alpha_rgb = [0.05, 0.06, 0.07]
- eigval = np.array([55.46, 4.794, 1.148])
- eigvec = np.array([[-0.5675, 0.7192, 0.4009],
- [-0.5808, -0.0045, -0.8140],
- [-0.5808, -0.0045, -0.8140]])
- f = AdjustLighting(alpha_rgb=alpha_rgb, eigval=eigval.ravel().tolist(), eigvec=eigvec.ravel().tolist())
- out_nd = f(nd.array(data_in, dtype=np.uint8))
- out_gt = np.clip(data_in.astype(np.float32)
- + np.dot(eigvec * alpha_rgb, eigval.reshape((3, 1))).reshape((1, 1, 3)), 0, 255).astype(np.uint8)
- assert_almost_equal(out_nd.asnumpy(), out_gt)
+ out_nd = transforms.ToTensor()(nd.array(data_in, dtype='uint8'))
+ assert_almost_equal(out_nd.asnumpy(), np.transpose(
+ data_in.astype(dtype=np.float32) / 255.0, (2, 0, 1)))
+
+def test_normalize():
+ data_in = np.random.uniform(0, 255, (300, 300, 3)).astype(dtype=np.uint8)
+ data_in = transforms.ToTensor()(nd.array(data_in, dtype='uint8'))
+ out_nd = transforms.Normalize(mean=(0, 1, 2), std=(3, 2, 1))(data_in)
+ data_expected = data_in.asnumpy()
+ data_expected[:][:][0] = data_expected[:][:][0] / 3.0
+ data_expected[:][:][1] = (data_expected[:][:][1] - 1.0) / 2.0
+ data_expected[:][:][2] = data_expected[:][:][2] - 2.0
+ assert_almost_equal(data_expected, out_nd.asnumpy())
+
+def test_flip_left_right():
+ data_in = np.random.uniform(0, 255, (300, 300, 3)).astype(dtype=np.uint8)
+ pil_img = Image.fromarray(data_in).transpose(Image.FLIP_LEFT_RIGHT)
+ data_trans = nd.image.flip_left_right(nd.array(data_in, dtype='uint8'))
+ assert_almost_equal(np.array(pil_img), data_trans.asnumpy())
+
+def test_flip_top_bottom():
+ data_in = np.random.uniform(0, 255, (300, 300, 3)).astype(dtype=np.uint8)
+ pil_img = Image.fromarray(data_in).transpose(Image.FLIP_TOP_BOTTOM)
+ data_trans = nd.image.flip_top_bottom(nd.array(data_in, dtype='uint8'))
+ assert_almost_equal(np.array(pil_img), data_trans.asnumpy())
if __name__ == '__main__':
import nose
--
To stop receiving notification emails like this one, please contact
"commits@mxnet.apache.org" <co...@mxnet.apache.org>.