Adaptive Data-Free Quantization [CVPR 2023]

This repository is the official code for the paper "Adaptive Data-Free Quantization" by Biao Qian, Yang Wang (corresponding author: yangwang@hfut.edu.cn), Richang Hong, Meng Wang (CVPR 2023, Vancouver, Canada).

Introduction

To address the overfitting and underfitting issues in data-free quantization (DFQ) task, we aim to study several critical questions --- how to measure the sample adaptability to Q under varied bit-width scenarios? whether the largest adaptability is the best? how to generate the samples with adaptive adapt- ability to improve Q’s generalization? To this end, we revisit DFQ from a zero-sum game perspective upon the sample adaptability between two players — a generator and a quantized network (Q). Following this viewpoint, we further define the disagreement and agreement samples to form two boundaries, where the margin between two boundaries is optimized to adaptively regulate the adaptability of generated samples to Q.

Dependencies

• python3.6
• pytorch1.3.1
• cuda10.0
• lmdb

Usages

On CIFAR-100, taking the pre-trained ResNet-20 as an example, run the following command:

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

On ImageNet, taking the pre-trained ResNet-18 as an example, run the following command:

python main.py --conf_path=./imagenet_res18.hocon --id=01

Results

The performance of our models is measured by Top-1 classification accuracy (%), which is reported below:

Citation

If you find the project codes useful for your research, please consider citing

@inproceedings{qian2023adaptive,
  title={Adaptive Data-Free Quantization},
  author={Qian, Biao and Wang, Yang and Hong, Richang and Wang, Meng},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={7960--7968},
  year={2023}
}