/AIB

Information Bottleneck Approach to Spatial Attention Learning, IJCAI2021

Primary LanguagePythonMIT LicenseMIT

Attentive Information Bottleneck

Overview

Pytorch code for "Information Bottleneck Approach to Spatial Attention Learning (IJCAI2021)".

What's in this repo so far:

  • Code for CIFAR-10/-100 experiments (VGG backbone) (this folder)
  • Code for CIFAR-10/-100 experiments (WRN backbone) (this folder)
  • Code for CUB experiments (VGG and WRN backbone) (this folder)

Reference Codes

[1] Attention Transfer

[2] Attention Branch Network

[3] LearnToPayAttention

Requirements

Create an anaconda environment:

$ conda env create -f environment.yaml

To run the code:

$ source activate torch36
$ <run_python_command> # see the examples in sub folders.

Citation

If you find this repository is useful, please cite the following reference.

@inproceedings{lai2021information,
    title = {Information Bottleneck Approach to Spatial Attention Learning},
    author = {Lai, Qiuxia and Li, Yu and Zeng, Ailing and Liu, Minhao and Sun, Hanqiu and Xu, Qiang},
    booktitle = {International Joint Conference on Artificial Intelligence},
    year = {2021}
}