tajanthan/pmf

Proximal Mean-field for Neural Network Quantization

Proximal Mean Field

This code implments the PMF algorithm described in the following paper:

"Proximal Mean-field for Neural Network Quantization", Thalaiyasingam Ajanthan, Puneet K. Dokania, Richard Hartley, and Philip H. S. Torr, ICCV, 2019.

Extensive hyperparamter tuning improved the classification accuracies compared to the original paper. The results are given below and the final hyperparameters can be found in hyperparams.txt.

Network	CIFAR-10	CIFAR-100	TinyImageNet
VGG-16	91.40%	64.71%	-
ResNet-18	93.24%	71.56%	51.52%

If you're using this code in a publication, please cite our paper.

@article{ajanthan2019pmf,
  author       = {Ajanthan, Thalaiyasingam and Dokania, Puneet K. and Hartley, Richard and Torr, Philip H. S.},
  title        = {Proximal Mean-field for Neural Network Quantization},
  journal      = {ICCV},
  year         = {2019}
}

This code is for research purposes only. If you want to use it for commercial purpose please contact us.

Contact: thalaiyasingam.ajanthan at anu.edu.au

Dependencies

• PyTorch, tested with PyTorch 1.0, and torchvision 0.2.1 in Python 2.
• numpy
• pandas
• bokeh

How to run the example

Example scripts on how to run the code can be found in the scripts/ folder.

bash scripts/cifar10-resnet18.bash

Acknowledgements

• BNN, for some utility functions.
• ResNet/VGG, for model definitions.