- Notifications
You must be signed in to change notification settings - Fork27
Code for “Discrimination-aware-Channel-Pruning-for-Deep-Neural-Networks”
License
NotificationsYou must be signed in to change notification settings
SCUT-AILab/DCP
Folders and files
| Name | Name | Last commit message | Last commit date | |
|---|---|---|---|---|
Repository files navigation
2019.05.10: We release a new version of dcp.
2018.11.27: We release the source code of dcp.
- python 2.7
- pytorch 0.4
- tensorflow
- pyhocon
- prettytable
Download the pre-trained pruned model from themodel zoo.
Add DCP into PYTHONPATH.
# This is my path of DCP. You need to change to your path of DCP.export PYTHONPATH=/home/liujing/Codes/Discrimination-aware-Channel-Pruning-for-Deep-Neural-Networks:$PYTHONPATH
- Set configuration for testing.You need to set
save_path,data_path,dataset,pruning_rate,net_type,depthand thepretrainedindcp/test/test.hocon.
cd dcp/test/vim test.hocon- Run testing.
python main.py test.hocon
- Download pre-trained mdoel.
- Add DCP into PYTHONPATH.
# This is my path of DCP. You need to change to your path of DCP.export PYTHONPATH=/home/liujing/Codes/Discrimination-aware-Channel-Pruning-for-Deep-Neural-Networks:$PYTHONPATH
- Before channel pruning, you need to add discrimination-aware loss and fine tune the whole network. You need to set
save_path,data_path,experiment_idand thepretrainedindcp/auxnet/cifar10_resnet.hocon.
cd dcp/auxnet/vim cifar10_resnet.hocon- Add discrimination-aware loss and conduct fine-tuning.
python main.py cifar10_resnet.hocon
- Set configuration for channel selection. You need to set
save_path,data_path,pruning_rateandexperiment_idindcp/channel_selection/cifar10_resnet.hocon. Additionally, you need to setpretrainedto the path ofbest_model_with_aux_fc.pthincheck_pointfolder.
cd dcp/channel_selection/vim cifar10_resnet.hocon- Conduct channel selection.
python main.py cifar10_resnet.hocon
- Set configuration for fine-tuning. You need to set
save_path,data_path, andexperiment_idindcp/finetune/cifar10_resnet.hocon. Additionally, you need to setpretrainedto the path ofmodel_xxx_cs_000.pthincheck_pointfolder.
cd dcp/finetune/vim cifar10_resnet.hocon- Fine-tune the pruned model.
python main cifar10_resnet.hocon
If you findDCP useful in your research, please consider to cite the following related papers:
@incollection{NIPS2018_7367,title = {Discrimination-aware Channel Pruning for Deep Neural Networks},author = {Zhuang, Zhuangwei and Tan, Mingkui and Zhuang, Bohan and Liu, Jing and Guo, Yong and Wu, Qingyao and Huang, Junzhou and Zhu, Jinhui},booktitle = {Advances in Neural Information Processing Systems 31},editor = {S. Bengio and H. Wallach and H. Larochelle and K. Grauman and N. Cesa-Bianchi and R. Garnett},pages = {881--892},year = {2018},publisher = {Curran Associates, Inc.},url = {http://papers.nips.cc/paper/7367-discrimination-aware-channel-pruning-for-deep-neural-networks.pdf}}About
Code for “Discrimination-aware-Channel-Pruning-for-Deep-Neural-Networks”
Resources
License
Stars
Watchers
Forks
Packages0
No packages published