PyTorch implementation of DAQ

This is an official implementation of the paper "Distance-aware Quantization", accepted to ICCV2021.

For more information, checkout the project site [website].

Getting started

Dependencies

• Python 3.6
• PyTorch = 1.5.0

Datasets

• Cifar-10
  • This can be automatically downloaded by learning our code
• ImageNet
  • This is available at here

Training & Evaluation

First, clone our github repository.

$ git clone https://github.com/cvlab-yonsei/DAQ.git

Cifar-10 dataset (ResNet-20 architecture)

• First, download full-precision model into results/ folder. Link: [weights]
• Note that you create results/ directory manually.

# Cifar-10 & ResNet-20 W1A1 model
$ python cifar10_train.py --config configs/DAQ/resnet20_DAQ_W1A1.yml
# Cifar-10 & ResNet-20 W1A32 model
$ python cifar10_train.py --config configs/DAQ/resnet20_DAQ_W1A32.yml

ImageNet dataset (ResNet-18 architecture)

• Will be released

# ImageNet & ResNet-18 W1A1 model
# ImageNet & ResNet-18 W1A1 model

Using the pretrained models

• ResNet-20
  • W1A1 [weights]
  • W1A32 [weights]
• ResNet-18
  • Will be released

Citation

@inproceedings{kim2021daq,
    author={Kim, Dohyung  and Lee, Junghyup and Ham, Bumsub},
    title={Distance-aware Quantization},
    booktitle={Proceedings of International Conference on Computer Vision},
    year={2021},
}

Credit

• ImageNet training code: [PyTorch official example code]
• ResNet-18 models: [PyTorch official code]
• ResNet-20 model: [ResNet on CIFAR10] [IRNet]
• Quantized modules: [DSQ]