liuguoyou / gdfq

official implementation of Generative Low-bitwidth Data Free Quantization(GDFQ)

Generative-Low-bitwidth-Data-Free-Quantization

We provide PyTorch implementation for "Generative Low bitwidth Data Free Quantization".

Paper

• Generative Low bitwidth Data Free Quantization
• Shoukai Xu ^*, Haokun Li ^*, Bohan Zhuang ^*, Jiezhang Cao, Jing Liu, Chuangrun Liang, Mingkui Tan ^*
• The European Conference on Computer Vision(ECCV), 2020

Dependencies

• Python 3.6
• PyTorch 1.2.0
• dependencies in requirements.txt

Getting Started

Installation

1. Clone this repo:

    git clone https://github.com/xushoukai/Generative-Low-bitwidth-Data-Free-Quantization.git
    cd Generative-Low-bitwidth-Data-Free-Quantization
   

2. Install pytorch and other dependencies.

   • pip install -r requirements.txt

Set the paths of datasets for testing

1. Set the "dataPath" in "cifar100_resnet20.hocon" as the path root of your CIFAR-100 dataset. For example:

   dataPath = "/home/datasets/Datasets/cifar"

2. Set the "dataPath" in "imagenet_resnet18.hocon" as the path root of your ImageNet dataset. For example:

   dataPath = "/home/datasets/Datasets/imagenet"

Training

To quantize the pretrained ResNet-20 on CIFAR-100 to 4-bit:

python main.py --conf_path ./cifar100_resnet20.hocon --id 01

To quantize the pretrained ResNet-18 on ImageNet to 4-bit:

python main.py --conf_path ./imagenet_resnet18.hocon --id 01

Results

Dataset	Model	Pretrain Top1 Acc(%)	W4A4(ours) Top1 Acc(%)
CIFAR-100	ResNet-20	70.33	63.58 ± 0.23
ImageNet	ResNet-18	71.47	60.60 ± 0.15

Note that we use the pretrained models from pytorchcv.

Citation

If this work is useful for your research, please cite our paper:

@InProceedings{xu2020generative,
title = {Multi-marginal Wasserstein GAN},
author = {Shoukai, Xu and Haokun, Li and Bohan, Zhuang and Jing, Liu and Jiezhang, Cao and Chuangrun, Liang and Mingkui, Tan},
booktitle = {The European Conference on Computer Vision},
year = {2020}
}

Acknowledgments

This work was partially supported by Key-Area Research and Development Program of Guangdong Province (2019B010155002, 2018B010107001, 2019B010155-001), National Natural Science Foundation of China(NSFC) 61836003 (key project), 2017ZT07X183, Tencent AI Lab Rhino-Bird Focused Research Program (No.JR201902), Fundamental Research Funds for the Central Universities D2191240.