[GitHub] ijkguo opened a new pull request #11373: update rcnn example

[GitBox <gi...@apache.org>]

ijkguo opened a new pull request #11373: update rcnn example
URL: https://github.com/apache/incubator-mxnet/pull/11373
 
 
   ## Description ##
   People complain about the Faster R-CNN example a lot.
   
   - Hard to install. Now we only require MXNet and pip packages. No Cython, CUDA involved anymore.
   - Too much code, some of them duplicate. Now it is reduced from ~9k to 3k lines. We dropped support for Fast R-CNN but we support end-to-end training. Why still maintain two-stage Fast R-CNN?
   - Too difficult to configure. The global config is gone. Now all hyper-parameters mentioned in the original paper is tunable through command line arguments. Just type `python3 train.py -h` to see all of them.
   - Not compatible with MXNet API changes. It usually involves trivial changes to api parameters but we reduced the code amount to make changes easier in the future. The old released models and training scripts are still working.
   
   ## Checklist ##
   ### Essentials ###
   - Changes are complete (i.e. I finished coding on this PR)
   - For new examples, README.md is added to explain the what the example does, the source of the dataset, expected performance on test set and reference to the original paper if applicable
   - To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change
   
   ### Changes ###
   All changes are limited to example/rcnn folder.
   - Reorganize the code to be dataset loading, data processing, network construction and user scripts.
   - Remove support of two-stage Fast R-CNN training so that the code is concise and fast.
   - Remove duplicate and unnecessary data processing code.
   - Remove shell scripts. It is much simpler to use to demo, train or test directly.
   
   ## Comments ##
   Note that data processing speed of the pure numpy code could be a limitation. However, numpy code is concise and accurate, which can serve as future reference to build better implementations.

----------------------------------------------------------------
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