Progressive Channel-Shrinking Network

This is the PyTorch implementation for "Progressive Channel-Shrinking (PCS) Network" by Jianhong Pan, Siyuan Yang, Lin Geng Foo, Qiuhong Ke, Hossein Rahmani, Zhipeng Fan, and Jun Liu

Citation

If you find our project useful in your research, please consider citing:

@ARTICLE{10169086,
  author={Pan, Jianhong and Yang, Siyuan and Foo, Lin Geng and Ke, Qiuhong and Rahmani, Hossein and Fan, Zhipeng and Liu, Jun},
  journal={IEEE Transactions on Multimedia}, 
  title={Progressive Channel-Shrinking Network}, 
  year={2023},
  volume={},
  number={},
  pages={1-11},
  doi={10.1109/TMM.2023.3291197}}

Introduction

Progressive Channel-Shrinking (PCS) can compress the selected salience entries at run-time instead of roughly approximating them to zero. Running Shrinking Policy provides a testing-static pruning scheme that can reduce the memory access cost for filter indexing. The evaluation of our method on ImageNet and CIFAR10 datasets over two prevalent networks: ResNet and VGG demonstrates that our PCS outperforms all baselines and achieves state-of-the-art in terms of compression- performance tradeoff. Moreover, it can be observed a significant and practical acceleration of inference.

Figure 1: Overview of a PCS layer.