SPEM: Self-adaptive Pooling Enhanced Attention Module for Image Recognition

This repository is the implementation of "SPEM: Self-adaptive Pooling Enhanced Attention Module for Image Recognition" [paper] on CIFAR-100 and CIFAR-10 datasets. Our paper has been accepted for presentation at MMM 2023. You can also check with the proceeding version.

Introduction

SPEM is a self-attention module. We empirically find and verify a phenomenon that the simple linear combination of global max-pooling and global min-pooling can produce pooling strategies that match or exceed the performance of global average pooling. Based on this empirical observation, we propose a simple-yet-effective self-attention module SPEM, which adopts a self-adaptive pooling strategy based on global max-pooling, global min-pooling and a lightweight module for producing the attention map.

Requirement

Python and PyTorch.

pip install -r requirements.txt

Usage

CUDA_VISIBLE_DEVICES=0 python run.py --dataset cifar10 --block-name bottleneck --depth 164 --epochs 164 --schedule 81 122 --gamma 0.1 --wd 1e-4

Results

	Dataset	original	SPEM
ResNet164	CIFAR10	93.39	94.80
ResNet164	CIFAR100	74.30	76.31

Citing SPEM

@inproceedings{zhong2023spem,
  title={SPEM: Self-adaptive Pooling Enhanced Attention Module for Image Recognition},
  author={Zhong, Shanshan and Wen, Wushao and Qin, Jinghui},
  booktitle={MultiMedia Modeling: 29th International Conference, MMM 2023, Bergen, Norway, January 9--12, 2023, Proceedings, Part II},
  pages={41--53},
  year={2023},
  organization={Springer}
}

Acknowledgments

Many thanks to bearpaw for his simple and clean Pytorch framework for image classification task.