Description

The implementation of "AACP: Model Compression by Accurate and Automatic Channel Pruning"(accepted by ICPR2022), by Lanbo Lin, Shengjie Chen, Yujiu Yang and Zhenhua Guo.

Table of Contents

1. Installation
2. Usage
3. Pretrained Models

Installation

Before running the code, installing the packages in requirements.txt. Note that the apex is required to be install by compiling the source code to avoid errors occur. You can install apex by:

git clone https://github.com/NVIDIA/apex.git
cd apex
pip install -v --no-cache-dir ./

Usage

We provide the code of pruning VGG16/Resnet56/Resnet110 on CIFAR10 and Resnet50/Mobilenetv2 on ImageNet (ILSVRC2012). All commands are run in the directory of " your_path/AACP"

cd your_path/AACP

CIFAR10

Pruning VGG16 on CIFAR10

1. Train an unpruned VGG16 model. Change the configurations and run

bash scripts/train_cifar10_vgg.sh

2. Search the optimal pruned architecture and finetune the searched result. Change the configurations and run

bash scripts/prune_cifar10_vgg.sh

Logs are saved to ./results by default.

Pruning Resnet56 on CIFAR10

1. Train an unpruned Resnet56 model. Change the configurations and run

bash scripts/train_cifar10_resnet56.sh

2. Search the optimal pruned architecture and finetune the searched result. Change the configurations and run

bash scripts/prune_cifar10_resnet56.sh

Pruning Resnet110 on CIFAR10

1. Train an unpruned Resnet56 model. Change the configurations and run

bash scripts/train_cifar10_resnet110.sh

2. Search the optimal pruned architecture and finetune the searched result. Change the configurations and run

bash scripts/prune_cifar10_resnet110.sh

ImageNet

Pruning ResNet50 on ImageNet

1. Train an unpruned Resnet50 model. Change the configurations and run

bash scripts/train_imagenet_resnet50.sh

2. Search the optimal pruned architecture. Change the configurations and run

bash scripts/prune_imagenet_resnet50.sh

3. Finetune the searched result. Change the configurations and run

bash scripts/finetune_imagenet_resnet50.sh

Pruning Mobilenetv2 on ImageNet

1. Train an unpruned mobilenetv2 model. Change the configurations and run

bash scripts/train_imagenet_mobilenetv2.sh

2. Search the optimal pruned architecture. Change the configurations and run

bash scripts/prune_imagenet_mobilenetv2.sh

3. Finetune the searched result. Change the configurations and run

bash scripts/finetune_imagenet_mobilenetv2.sh

Pretrained Models

Coming soon.