Awesome

WHC: Weighted Hybrid Criterion for Filter Pruning on Convolutional Neural Networks

(ICASSP 2023)

Please cite our work as follows:

@inproceedings{WHC,
  author    = {Shaowu Chen and Weize Sun and Lei Huang},
  title     = {WHC: Weighted Hybrid Criterion for Filter Pruning on Convolutional Neural Networks},
  booktitle = {ICASSP},
  year      = {2023}
}

The implementaion is based on FPGM. Thanks to YangHe for his help and contribution.

1. Environment:

python3.6.12 ; Torch 1.3.1.

2. Description for files:

  ├── pruning_cifar10_orig.py: Code for CIFAR-10
  ├── pruning_imagenet.py: Code for ImageN
  ├── run.sh: Script demo to run the code
  ├── utils.py 
  ├── models

3. Log files and CKPT:

Find log files and checkpoints in WHC Google Drive.

Find pre-trained CIFAR-10 parameters (unpruned) in FPGM Google Drive.

Find Pytorch official pre-trained ImageNet parameters (unpruned) in resnet18, resnet34, resnet50, resnet101.