Channel Equilibrium Networks for Learning Deep Representation

This is the PyTorch implementation of the paper Channel Equilibrium Networks for Learning Deep Representation, ICML2020.

By Wenqi Shao, Shitao Tang, Xingang Pan, Ping Tan, Xiaogang Wang, Ping Luo.

[https://arxiv.org/pdf/2003.00214.pdf)

Channel Equilibrium Networks

We design a novel neural building block, termed Channel Equilibrium (CE) block, which enables channels at the same layer to contribute equally to the learned representation.

Comparisons of top-1 accuracies on the validation set of ImageNet, by using ResNet50 trained with BN and CE.

Model	Top-1	Top-5
ResNet50-BN	76.6	93.0
ResNet50-CE	78.2	94.1

Getting Started

• Install PyTorch
• Clone the repo:

  git clone https://github.com/Tangshitao/CENet.git
  

Requirements

• python packages
  • pytorch>=0.4.0
  • torchvision>=0.2.1
  • tensorboardX
  • pyyaml

Data Preparation

• Download the ImageNet dataset and put them into the {repo_root}/data/imagenet.

Training a model from scratch

./train.sh

Number of GPUs and configuration file to use can be modified in train.sh

Evaluating performance of a model

Download the pretrained models from Model Zoo and put them into the {repo_root}/model_zoo

./test.sh

Or you can specify the checkpoint path by modifying test.sh

--checkpoint_path model_zoo/ssn_8x2_75.848.pth \

Model Zoo

We provide models pretrained with CE block on ImageNet.

Model	Top-1*	Top-5*	Download
ResNet50v1+CE	78.2%	94.1%	[Google Drive]

*single-crop validation accuracy on ImageNet (a 224x224 center crop from resized image with shorter side=256)

In evaluation, download the above models and put them into the {repo_root}/model_zoo.

Citation

If you find this work helpful in your project or use our model zoo, please consider citing:

@article{shao2020channel,
  title={Channel equilibrium networks for learning deep representation},
  author={Shao, Wenqi and Tang, Shitao and Pan, Xingang and Tan, Ping and Wang, Xiaogang and Luo, Ping},
  journal={arXiv preprint arXiv:2003.00214},
  year={2020}
}