Filter-Pruning-via-KMEANS-Clustering

Pytorch implementation for "More similar Less Important: Filter Pruning via KMeans Clustering" IEEE ICME2021

This implementation is based on filter-pruning-geometric-median.

Requirements

• Python 3.6
• PyTorch 0.3.1
• TorchVision 0.3.0

Training ResNet on ImageNet

Usage of Pruning Training

We train each model from scratch with stepwise learning rate decay by default. If you wish to train the model with pre-trained models and cosine annealing learning rate decay, please use the options --use_pretrain --lr 0.01 --cos

Run Pruning Training ResNet (depth 101,50,34,18) on Imagenet:

python pruning_kmeans_imagenet.py -a resnet101 --save_dir ./snapshots/resnet101_8_04 --pruning_rate 0.4 --n_clusters 8 --layer_begin 0 --layer_end 309 --layer_inter 3  /path/to/Imagenet2012

python pruning_kmeans_imagenet.py -a resnet50  --save_dir ./snapshots/resnet50_8_04 --pruning_rate 0.4 --n_clusters 8 --layer_begin 0 --layer_end 156 --layer_inter 3  /path/to/Imagenet2012

python pruning_kmeans_imagenet.py -a resnet34  --save_dir ./snapshots/resnet34_8_04 --pruning_rate 0.4 --n_clusters 8 --layer_begin 0 --layer_end 105 --layer_inter 3  /path/to/Imagenet2012

python ppruning_kmeans_imagenet.py -a resnet18  --save_dir ./snapshots/resnet18_8_04 --pruning_rate 0.4 --n_clusters 8 --layer_begin 0 --layer_end 57 --layer_inter 3  /path/to/Imagenet2012

Training ResNet on CIFAR10

Usage of Pruning Training

sh scripts/pruning_cifar10_new.sh