Boundary-aware Graph Reasoning for Semantic Segmentation

TL;DR

This paper introduces a Boundary-aware Graph Reasoning (BGR) module that enhances semantic segmentation by focusing on boundary regions using boundary scores, improving contextual feature learning with efficient graph convolution.

Contribution

The paper presents a novel BGR module that leverages boundary scores to guide graph reasoning, effectively improving segmentation accuracy with reduced computational cost.

Findings

BGR improves segmentation accuracy on three benchmarks.

Using boundary scores enhances focus on boundary regions.

Efficient graph convolution reduces computational overhead.

Abstract

In this paper, we propose a Boundary-aware Graph Reasoning (BGR) module to learn long-range contextual features for semantic segmentation. Rather than directly construct the graph based on the backbone features, our BGR module explores a reasonable way to combine segmentation erroneous regions with the graph construction scenario. Motivated by the fact that most hard-to-segment pixels broadly distribute on boundary regions, our BGR module uses the boundary score map as prior knowledge to intensify the graph node connections and thereby guide the graph reasoning focus on boundary regions. In addition, we employ an efficient graph convolution implementation to reduce the computational cost, which benefits the integration of our BGR module into current segmentation backbones. Extensive experiments on three challenging segmentation benchmarks demonstrate the effectiveness of our proposed BGR module for semantic segmentation.