liuchunlei0430 / cbcn_pytorch Goto Github PK

Environment: PyTorch (0.4.0), torchvision (0.2.1), tensorboardX, python3, CUDA(8.0)
• git clone https://github.com/liuchunlei0430/CBCN_Pytorch.git get the ImageNet dataset ready
Install Convolutional Module and Binary Module
• cd install
• sh install.sh
• cd BinActivateFunc_PyTorch
• sh install.sh

• Train: python ImageNet.py [dataset_dir] --tensorboard
• Evaluation: python ImageNet.py [dataset_dir] --pretrained --tensorboard
CBCN(without centerloss finetune) models can be obtained in https://drive.google.com/open?id=1wXL5LJAE6oduDBM7if4zeoSIJ0tjS4sS.
CBCN(with centerloss finetune) is prepared to upload in Google Drive.

Thanks for the code of ORN! Inspired by ORN which already show their powerful ability in within-class rotation-invariance, we also employ similar way to enhance the representative ability which destoryed by the binarization process. We focus on storate reduction in our CBCNs. More detail can be seen in the install/orn/modules/ORConv.py.

@inproceedings{liu2019cbcn,
title={Circulant Binary Convolutional Networks: Enhancing the Performance of 1-bit DCNNs with Circulant Back Propagation},
author={Liu, ChunLei and Ding, Wenrui and Xia, Xin and Zhang, Baochang and Gu, Jiaxin and Liu, Jianzhuang and Ji, Rongrong and David, Doermann },
booktitle={IEEE Conference on Computer Vision and Pattern Recognition},
year={2019}
}
@ inproceedings{Zhou2017ORN,
author = {Zhou, Yanzhao and Ye, Qixiang and Qiu, Qiang and Jiao, Jianbin},
title = {Oriented Response Networks},
booktitle = {CVPR},
year = {2017}
}