You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@mxnet.apache.org by GitBox <gi...@apache.org> on 2019/01/11 14:05:43 UTC
[GitHub] TaoLv closed pull request #12498: Code modification for testcases
of various network models in directory example
TaoLv closed pull request #12498: Code modification for testcases of various network models in directory example
URL: https://github.com/apache/incubator-mxnet/pull/12498
This is a PR merged from a forked repository.
As GitHub hides the original diff on merge, it is displayed below for
the sake of provenance:
As this is a foreign pull request (from a fork), the diff is supplied
below (as it won't show otherwise due to GitHub magic):
diff --git a/example/bayesian-methods/README.md b/example/bayesian-methods/README.md
index ec9e8be8692..fc35b94219d 100644
--- a/example/bayesian-methods/README.md
+++ b/example/bayesian-methods/README.md
@@ -11,3 +11,27 @@ and *Bayesian Dark Knowledge (BDK)* [<cite>(Balan, Rathod, Murphy and Welling, 2
**bdk.ipynb** shows how to use MXNet to implement the DistilledSGLD algorithm in Bayesian Dark Knowledge.
**bdk_demo.py** contains scripts (more than the notebook) related to Bayesian Dark Knowledge. Use `python bdk_demo.py -d 1 -l 2 -t 50000` to run classification on MNIST.
+
+View parameters we can use with the following command.
+
+```shell
+python bdk_demo.py -h
+
+
+usage: bdk_demo.py [-h] [-d DATASET] [-l ALGORITHM] [-t TRAINING] [--gpu GPU]
+
+Examples in the paper [NIPS2015]Bayesian Dark Knowledge and [ICML2011]Bayesian
+Learning via Stochastic Gradient Langevin Dynamics
+
+optional arguments:
+ -h, --help show this help message and exit
+ -d DATASET, --dataset DATASET
+ Dataset to use. 0 --> TOY, 1 --> MNIST, 2 -->
+ Synthetic Data in the SGLD paper
+ -l ALGORITHM, --algorithm ALGORITHM
+ Type of algorithm to use. 0 --> SGD, 1 --> SGLD,
+ other-->DistilledSGLD
+ -t TRAINING, --training TRAINING
+ Number of training samples
+ --gpu GPU 0 to use GPU, not set to use CPU
+```
diff --git a/example/bayesian-methods/bdk_demo.py b/example/bayesian-methods/bdk_demo.py
index 145dac10e2a..cd39bfd2a7c 100644
--- a/example/bayesian-methods/bdk_demo.py
+++ b/example/bayesian-methods/bdk_demo.py
@@ -156,34 +156,34 @@ def get_toy_sym(teacher=True, teacher_noise_precision=None):
return net
-def dev():
- return mx.gpu()
+def dev(gpu_id=None):
+ return mx.gpu(gpu_id) if gpu_id else mx.cpu()
-def run_mnist_SGD(training_num=50000):
+def run_mnist_SGD(training_num=50000, gpu_id=None):
X, Y, X_test, Y_test = load_mnist(training_num)
minibatch_size = 100
net = get_mnist_sym()
data_shape = (minibatch_size,) + X.shape[1::]
- data_inputs = {'data': nd.zeros(data_shape, ctx=dev()),
- 'softmax_label': nd.zeros((minibatch_size,), ctx=dev())}
+ data_inputs = {'data': nd.zeros(data_shape, ctx=dev(gpu_id)),
+ 'softmax_label': nd.zeros((minibatch_size,), ctx=dev(gpu_id))}
initializer = mx.init.Xavier(factor_type="in", magnitude=2.34)
- exe, exe_params, _ = SGD(sym=net, dev=dev(), data_inputs=data_inputs, X=X, Y=Y,
+ exe, exe_params, _ = SGD(sym=net, dev=dev(gpu_id), data_inputs=data_inputs, X=X, Y=Y,
X_test=X_test, Y_test=Y_test,
total_iter_num=1000000,
initializer=initializer,
lr=5E-6, prior_precision=1.0, minibatch_size=100)
-def run_mnist_SGLD(training_num=50000):
+def run_mnist_SGLD(training_num=50000, gpu_id=None):
X, Y, X_test, Y_test = load_mnist(training_num)
minibatch_size = 100
net = get_mnist_sym()
data_shape = (minibatch_size,) + X.shape[1::]
- data_inputs = {'data': nd.zeros(data_shape, ctx=dev()),
- 'softmax_label': nd.zeros((minibatch_size,), ctx=dev())}
+ data_inputs = {'data': nd.zeros(data_shape, ctx=dev(gpu_id)),
+ 'softmax_label': nd.zeros((minibatch_size,), ctx=dev(gpu_id))}
initializer = mx.init.Xavier(factor_type="in", magnitude=2.34)
- exe, sample_pool = SGLD(sym=net, dev=dev(), data_inputs=data_inputs, X=X, Y=Y,
+ exe, sample_pool = SGLD(sym=net, dev=dev(gpu_id), data_inputs=data_inputs, X=X, Y=Y,
X_test=X_test, Y_test=Y_test,
total_iter_num=1000000,
initializer=initializer,
@@ -191,7 +191,7 @@ def run_mnist_SGLD(training_num=50000):
thin_interval=100, burn_in_iter_num=1000)
-def run_mnist_DistilledSGLD(training_num=50000):
+def run_mnist_DistilledSGLD(training_num=50000, gpu_id=None):
X, Y, X_test, Y_test = load_mnist(training_num)
minibatch_size = 100
if training_num >= 10000:
@@ -214,10 +214,10 @@ def run_mnist_DistilledSGLD(training_num=50000):
logsoftmax = LogSoftmax()
student_net = get_mnist_sym(output_op=logsoftmax, num_hidden=num_hidden)
data_shape = (minibatch_size,) + X.shape[1::]
- teacher_data_inputs = {'data': nd.zeros(data_shape, ctx=dev()),
- 'softmax_label': nd.zeros((minibatch_size,), ctx=dev())}
- student_data_inputs = {'data': nd.zeros(data_shape, ctx=dev()),
- 'softmax_label': nd.zeros((minibatch_size, 10), ctx=dev())}
+ teacher_data_inputs = {'data': nd.zeros(data_shape, ctx=dev(gpu_id)),
+ 'softmax_label': nd.zeros((minibatch_size,), ctx=dev(gpu_id))}
+ student_data_inputs = {'data': nd.zeros(data_shape, ctx=dev(gpu_id)),
+ 'softmax_label': nd.zeros((minibatch_size, 10), ctx=dev(gpu_id))}
teacher_initializer = BiasXavier(factor_type="in", magnitude=1)
student_initializer = BiasXavier(factor_type="in", magnitude=1)
student_exe, student_params, _ = \
@@ -231,17 +231,17 @@ def run_mnist_DistilledSGLD(training_num=50000):
teacher_learning_rate=teacher_learning_rate,
student_learning_rate=student_learning_rate,
teacher_prior_precision=teacher_prior, student_prior_precision=student_prior,
- perturb_deviation=perturb_deviation, minibatch_size=100, dev=dev())
+ perturb_deviation=perturb_deviation, minibatch_size=100, dev=dev(gpu_id))
-def run_toy_SGLD():
+def run_toy_SGLD(gpu_id=None):
X, Y, X_test, Y_test = load_toy()
minibatch_size = 1
teacher_noise_precision = 1.0 / 9.0
net = get_toy_sym(True, teacher_noise_precision)
data_shape = (minibatch_size,) + X.shape[1::]
- data_inputs = {'data': nd.zeros(data_shape, ctx=dev()),
- 'teacher_output_label': nd.zeros((minibatch_size, 1), ctx=dev())}
+ data_inputs = {'data': nd.zeros(data_shape, ctx=dev(gpu_id)),
+ 'teacher_output_label': nd.zeros((minibatch_size, 1), ctx=dev(gpu_id))}
initializer = mx.init.Uniform(0.07)
exe, params, _ = \
SGLD(sym=net, data_inputs=data_inputs,
@@ -253,20 +253,20 @@ def run_toy_SGLD():
burn_in_iter_num=1000,
thin_interval=10,
task='regression',
- minibatch_size=minibatch_size, dev=dev())
+ minibatch_size=minibatch_size, dev=dev(gpu_id))
-def run_toy_DistilledSGLD():
+def run_toy_DistilledSGLD(gpu_id=None):
X, Y, X_test, Y_test = load_toy()
minibatch_size = 1
teacher_noise_precision = 1.0
teacher_net = get_toy_sym(True, teacher_noise_precision)
student_net = get_toy_sym(False)
data_shape = (minibatch_size,) + X.shape[1::]
- teacher_data_inputs = {'data': nd.zeros(data_shape, ctx=dev()),
- 'teacher_output_label': nd.zeros((minibatch_size, 1), ctx=dev())}
- student_data_inputs = {'data': nd.zeros(data_shape, ctx=dev())}
- # 'softmax_label': nd.zeros((minibatch_size, 10), ctx=dev())}
+ teacher_data_inputs = {'data': nd.zeros(data_shape, ctx=dev(gpu_id)),
+ 'teacher_output_label': nd.zeros((minibatch_size, 1), ctx=dev(gpu_id))}
+ student_data_inputs = {'data': nd.zeros(data_shape, ctx=dev(gpu_id))}
+
teacher_initializer = mx.init.Uniform(0.07)
student_initializer = mx.init.Uniform(0.07)
student_grad_f = lambda student_outputs, teacher_pred: \
@@ -284,21 +284,21 @@ def run_toy_DistilledSGLD():
student_grad_f=student_grad_f,
teacher_prior_precision=0.1, student_prior_precision=0.001,
perturb_deviation=0.1, minibatch_size=minibatch_size, task='regression',
- dev=dev())
+ dev=dev(gpu_id))
-def run_toy_HMC():
+def run_toy_HMC(gpu_id=None):
X, Y, X_test, Y_test = load_toy()
minibatch_size = Y.shape[0]
noise_precision = 1 / 9.0
net = get_toy_sym(True, noise_precision)
data_shape = (minibatch_size,) + X.shape[1::]
- data_inputs = {'data': nd.zeros(data_shape, ctx=dev()),
- 'teacher_output_label': nd.zeros((minibatch_size, 1), ctx=dev())}
+ data_inputs = {'data': nd.zeros(data_shape, ctx=dev(gpu_id)),
+ 'teacher_output_label': nd.zeros((minibatch_size, 1), ctx=dev(gpu_id))}
initializer = mx.init.Uniform(0.07)
sample_pool = HMC(net, data_inputs=data_inputs, X=X, Y=Y, X_test=X_test, Y_test=Y_test,
sample_num=300000, initializer=initializer, prior_precision=1.0,
- learning_rate=1E-3, L=10, dev=dev())
+ learning_rate=1E-3, L=10, dev=dev(gpu_id))
def run_synthetic_SGLD():
@@ -350,21 +350,22 @@ def run_synthetic_SGLD():
help="Type of algorithm to use. 0 --> SGD, 1 --> SGLD, other-->DistilledSGLD")
parser.add_argument("-t", "--training", type=int, default=50000,
help="Number of training samples")
+ parser.add_argument("--gpu", type=int, help="0 to use GPU, not set to use CPU")
args = parser.parse_args()
training_num = args.training
if args.dataset == 1:
if 0 == args.algorithm:
- run_mnist_SGD(training_num)
+ run_mnist_SGD(training_num, gpu_id=args.gpu)
elif 1 == args.algorithm:
- run_mnist_SGLD(training_num)
+ run_mnist_SGLD(training_num, gpu_id=args.gpu)
else:
- run_mnist_DistilledSGLD(training_num)
+ run_mnist_DistilledSGLD(training_num, gpu_id=args.gpu)
elif args.dataset == 0:
if 1 == args.algorithm:
- run_toy_SGLD()
+ run_toy_SGLD(gpu_id=args.gpu)
elif 2 == args.algorithm:
- run_toy_DistilledSGLD()
+ run_toy_DistilledSGLD(gpu_id=args.gpu)
elif 3 == args.algorithm:
- run_toy_HMC()
+ run_toy_HMC(gpu_id=args.gpu)
else:
run_synthetic_SGLD()
diff --git a/example/fcn-xs/README.md b/example/fcn-xs/README.md
index 145aa31cb70..49c57fc08ea 100644
--- a/example/fcn-xs/README.md
+++ b/example/fcn-xs/README.md
@@ -40,14 +40,33 @@ this is the fully convolution style of the origin
Once you completed all these steps, your working directory should contain a ```.\VOC2012``` directory, which contains the following: ```JPEGImages folder```, ```SegmentationClass folder```, ```train.lst```, ```val.lst```
#### Step 3: Train the fcn-xs model
-* Based on your hardware, configure GPU or CPU for training in `fcn_xs.py`. It is recommended to use GPU due to the computational complexity and data load.
-```python
-# ctx = mx.cpu(0)
-ctx = mx.gpu(0)
+* Based on your hardware, configure CPU or GPU for training by parameter ```--gpu```. It is recommended to use GPU due to the computational complexity and data load.
+View parameters we can use with the following command.
+```shell
+python fcn_xs.py -h
+
+
+usage: fcn_xs.py [-h] [--model MODEL] [--prefix PREFIX] [--epoch EPOCH]
+ [--init-type INIT_TYPE] [--retrain] [--gpu GPU]
+
+Convert vgg16 model to vgg16fc model.
+
+optional arguments:
+ -h, --help show this help message and exit
+ --model MODEL The type of fcn-xs model, e.g. fcnxs, fcn16s, fcn8s.
+ --prefix PREFIX The prefix(include path) of vgg16 model with mxnet
+ format.
+ --epoch EPOCH The epoch number of vgg16 model.
+ --init-type INIT_TYPE
+ the init type of fcn-xs model, e.g. vgg16, fcnxs
+ --retrain true means continue training.
+ --gpu GPU 0 to use GPU, not set to use CPU
```
+
* It is recommended to train fcn-32s and fcn-16s before training the fcn-8s model
To train the fcn-32s model, run the following:
+
```shell
python -u fcn_xs.py --model=fcn32s --prefix=VGG_FC_ILSVRC_16_layers --epoch=74 --init-type=vgg16
```
diff --git a/example/fcn-xs/fcn_xs.py b/example/fcn-xs/fcn_xs.py
index 53244a1759c..5b799f32e46 100644
--- a/example/fcn-xs/fcn_xs.py
+++ b/example/fcn-xs/fcn_xs.py
@@ -28,9 +28,10 @@
logger = logging.getLogger()
logger.setLevel(logging.INFO)
-ctx = mx.gpu(0)
+
def main():
+ ctx = mx.cpu() if not args.gpu else mx.gpu(args.gpu)
fcnxs = symbol_fcnxs.get_fcn32s_symbol(numclass=21, workspace_default=1536)
fcnxs_model_prefix = "model_pascal/FCN32s_VGG16"
if args.model == "fcn16s":
@@ -85,6 +86,7 @@ def main():
help='the init type of fcn-xs model, e.g. vgg16, fcnxs')
parser.add_argument('--retrain', action='store_true', default=False,
help='true means continue training.')
+ parser.add_argument("--gpu", type=int, help="0 to use GPU, not set to use CPU")
args = parser.parse_args()
logging.info(args)
main()
diff --git a/example/rcnn/README.md b/example/rcnn/README.md
index b5284183d16..5e6127ccb08 100644
--- a/example/rcnn/README.md
+++ b/example/rcnn/README.md
@@ -9,7 +9,7 @@ For a gluon imperative version, checkout https://github.com/dmlc/gluon-cv.
### Out-of-box inference models
Download any of the following models to the current directory and run `python3 demo.py --dataset $Dataset$ --network $Network$ --params $MODEL_FILE$ --image $YOUR_IMAGE$` to get single image inference.
-For example `python3 demo.py --dataset voc --network vgg16 --params vgg16_voc0712.params --image myimage.jpg`, add `--gpu 0` to use GPU optionally.
+For example `python3 demo.py --dataset voc --network vgg16 --params vgg16_voc0712.params --image myimage.jpg`, add `--gpu 0` to use GPU, not set to use CPU.
Different network has different configuration. Different dataset has different object class names. You must pass them explicitly as command line arguments.
| Network | Dataset | Imageset | Reference | Result | Link |
diff --git a/example/rcnn/demo.py b/example/rcnn/demo.py
index 2315bb8af36..b0a4ddbeab4 100644
--- a/example/rcnn/demo.py
+++ b/example/rcnn/demo.py
@@ -92,7 +92,7 @@ def parse_args():
parser.add_argument('--params', type=str, default='', help='path to trained model')
parser.add_argument('--dataset', type=str, default='voc', help='training dataset')
parser.add_argument('--image', type=str, default='', help='path to test image')
- parser.add_argument('--gpu', type=str, default='', help='gpu device eg. 0')
+ parser.add_argument('--gpu', type=str, default='', help='GPU devices, eg."0,1,2,3" , not set to use CPU.')
parser.add_argument('--vis', action='store_true', help='display results')
parser.add_argument('--vis-thresh', type=float, default=0.7, help='threshold display boxes')
# faster rcnn params
diff --git a/example/rcnn/test.py b/example/rcnn/test.py
index 3c047d22201..e964c908066 100644
--- a/example/rcnn/test.py
+++ b/example/rcnn/test.py
@@ -35,7 +35,7 @@ def test_net(sym, imdb, args):
logger.info('called with args\n{}'.format(pprint.pformat(vars(args))))
# setup context
- ctx = mx.gpu(args.gpu)
+ ctx = mx.cpu() if not args.gpu else mx.gpu(args.gpu)
# load testing data
test_data = TestLoader(imdb.roidb, batch_size=1, short=args.img_short_side, max_size=args.img_long_side,
@@ -94,7 +94,7 @@ def parse_args():
parser.add_argument('--params', type=str, default='', help='path to trained model')
parser.add_argument('--dataset', type=str, default='voc', help='training dataset')
parser.add_argument('--imageset', type=str, default='', help='imageset splits')
- parser.add_argument('--gpu', type=int, default=0, help='gpu device eg. 0')
+ parser.add_argument('--gpu', type=int, default=0, help='0 to use GPU, not set to use CPU')
# faster rcnn params
parser.add_argument('--img-short-side', type=int, default=600)
parser.add_argument('--img-long-side', type=int, default=1000)
diff --git a/example/rcnn/train.py b/example/rcnn/train.py
index 0739069afb4..7b1f2f7f31a 100644
--- a/example/rcnn/train.py
+++ b/example/rcnn/train.py
@@ -33,7 +33,7 @@ def train_net(sym, roidb, args):
logger.info('called with args\n{}'.format(pprint.pformat(vars(args))))
# setup multi-gpu
- ctx = [mx.gpu(int(i)) for i in args.gpus.split(',')]
+ ctx = [mx.cpu()] if not args.gpus else [mx.gpu(int(i)) for i in args.gpus.split(',')]
batch_size = args.rcnn_batch_size * len(ctx)
# load training data
@@ -127,7 +127,7 @@ def parse_args():
parser.add_argument('--pretrained', type=str, default='', help='path to pretrained model')
parser.add_argument('--dataset', type=str, default='voc', help='training dataset')
parser.add_argument('--imageset', type=str, default='', help='imageset splits')
- parser.add_argument('--gpus', type=str, default='0', help='gpu devices eg. 0,1')
+ parser.add_argument('--gpus', type=str, help='GPU devices, eg: "0,1,2,3" , not set to use CPU')
parser.add_argument('--epochs', type=int, default=10, help='training epochs')
parser.add_argument('--lr', type=float, default=0.001, help='base learning rate')
parser.add_argument('--lr-decay-epoch', type=str, default='7', help='epoch to decay lr')
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
With regards,
Apache Git Services