Structured Binary Neural Networks for Image Recognition

created by Bohan Zhuang

External code

• We use the PyTorch-Encoding as the code base. Please refer to Pytorch-Encoding for detailed instructions of installation and usage. Note that you must install pytorch 1.0 from source and you need to solve the bug via this link.

Training

• run python train.py. We use the binary backbone pretrained on ImageNet as initialization. The pretrained binary backbone can be downloaded at image classification repository.

• In particular, to make consistent with the conference version, we employ the learnt "scales" by default. For example, download the resnet18_with_softgates.pth.tar as initialization. But in the journal version, we omit these scales and directly use averaging.

Testing

python test.py

Citations

@article{zhuang2019structured,
  title={Structured Binary Neural Networks for Image Recognition},
  author={Zhuang, Bohan and Shen, Chunhua and Tan, Mingkui and Liu, Lingqiao and Reid, Ian},
  journal={arXiv preprint arXiv:1909.09934},
  year={2019}
}

License

For academic use, this project is licensed under the 2-clause BSD License - see the LICENSE file for details. For commercial use, please contact the authors.