FcaNet: Frequency Channel Attention Networks

TL;DR

FcaNet introduces a frequency domain approach to channel attention, improving image classification and detection performance by capturing richer information without increasing model complexity.

Contribution

The paper proposes multi-spectral channel attention using frequency analysis, generalizing global average pooling and enhancing existing attention mechanisms.

Findings

Achieves state-of-the-art results on classification, detection, and segmentation.

Outperforms SENet with same parameters and computational cost.

Demonstrates effectiveness of frequency domain analysis in attention mechanisms.

Abstract

Attention mechanism, especially channel attention, has gained great success in the computer vision field. Many works focus on how to design efficient channel attention mechanisms while ignoring a fundamental problem, i.e., channel attention mechanism uses scalar to represent channel, which is difficult due to massive information loss. In this work, we start from a different view and regard the channel representation problem as a compression process using frequency analysis. Based on the frequency analysis, we mathematically prove that the conventional global average pooling is a special case of the feature decomposition in the frequency domain. With the proof, we naturally generalize the compression of the channel attention mechanism in the frequency domain and propose our method with multi-spectral channel attention, termed as FcaNet. FcaNet is simple but effective. We can change a few lines of code in the calculation to implement our method within existing channel attention methods. Moreover, the proposed method achieves state-of-the-art results compared with other channel attention methods on image classification, object detection, and instance segmentation tasks. Our method could consistently outperform the baseline SENet, with the same number of parameters and the same computational cost. Our code and models will are publicly available at https://github.com/cfzd/FcaNet.