Compress networks using PyTorch - Pruning and Quantization

This is a complete training example for Deep Convolutional Networks on ImageNet.

Currently, the compression methods based on several techniques below:

• Taylor Expansion (A good summary of this approach can be found here).
• Attention Transfer from paper "Paying More Attention to Attention: Improving the Performance of Convolutional Neural Networks via Attention Transfer" (CVPR 2017)
• Knowledge Distillation from paper "Distilling the Knowledge in a Neural Network" (NIPS 2014)
• Quantization (grab some code from here)

Dependencies

• Python 3.6.3
• Pytorch 1.0

To clone:

git clone https://github.com/Yifan122/network_compress

example for pruning resnet network:

python resnet_prune.py --train_path /home/to/imagenet/taining/dataset --val_path /home/to/imagenet/validation/dataset