cvlab-yonsei / EWGS

An official implementation of "Network Quantization with Element-wise Gradient Scaling" (CVPR 2021) in PyTorch.

PyTorch implementation of EWGS

This is the implementation of the paper "Network Quantization with Element-wise Gradient Scaling".

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

Requirements

• Python >= 3.6
• PyTorch >= 1.3.0

Datasets

• CIFAR-10 (will be automatically downloaded when you run the code)
• ImageNet (ILSVRC-2012) available at http://www.image-net.org

Code

Please refer to the run.sh files in the CIFAR10 and ImageNet folders.

Bibtex

@inproceedings{lee2021network,
  title={Network Quantization with Element-wise Gradient Scaling},
  author={Lee, Junghyup and Kim, Dohyung and Ham, Bumsub},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  year={2021}
}

References

• ImageNet training code: [PyTorch official example code]
• ResNet-18/34 models: [PyTorch official code]
• ResNet-20 model: [ResNet on CIFAR10] [IRNet]
• Quantized modules: [DSQ]
• Estimating Hessian trace: [PyHessian]