Low-Resolution Self-Attention for Semantic Segmentation

TL;DR

This paper introduces Low-Resolution Self-Attention (LRSA), a novel mechanism that captures global context efficiently for semantic segmentation by computing self-attention at a fixed low resolution, reducing computational costs while maintaining high performance.

Contribution

The paper proposes LRSA, a new self-attention mechanism that reduces computational complexity in vision transformers for segmentation tasks, and demonstrates its effectiveness with the LRFormer model.

Findings

LRFormer outperforms state-of-the-art models on multiple datasets.

LRSA significantly reduces FLOPs compared to traditional high-resolution self-attention.

The approach maintains high segmentation accuracy with lower computational cost.

Abstract

Semantic segmentation tasks naturally require high-resolution information for pixel-wise segmentation and global context information for class prediction. While existing vision transformers demonstrate promising performance, they often utilize high-resolution context modeling, resulting in a computational bottleneck. In this work, we challenge conventional wisdom and introduce the Low-Resolution Self-Attention (LRSA) mechanism to capture global context at a significantly reduced computational cost, i.e., FLOPs. Our approach involves computing self-attention in a fixed low-resolution space regardless of the input image's resolution, with additional 3x3 depth-wise convolutions to capture fine details in the high-resolution space. We demonstrate the effectiveness of our LRSA approach by building the LRFormer, a vision transformer with an encoder-decoder structure. Extensive experiments on the ADE20K, COCO-Stuff, and Cityscapes datasets demonstrate that LRFormer outperforms state-of-the-art models. Code is available at https://github.com/yuhuan-wu/LRFormer.