ExFuse: Enhancing Feature Fusion for Semantic Segmentation

TL;DR

ExFuse introduces a novel feature fusion framework for semantic segmentation that effectively bridges the semantic and resolution gap between features, leading to significant performance improvements.

Contribution

The paper proposes ExFuse, a new framework that enhances feature fusion by integrating semantic info into low-level features and details into high-level features.

Findings

Achieves 87.9% mean IoU on PASCAL VOC 2012

Improves segmentation quality by 4.0%

Outperforms previous state-of-the-art methods

Abstract

Modern semantic segmentation frameworks usually combine low-level and high-level features from pre-trained backbone convolutional models to boost performance. In this paper, we first point out that a simple fusion of low-level and high-level features could be less effective because of the gap in semantic levels and spatial resolution. We find that introducing semantic information into low-level features and high-resolution details into high-level features is more effective for the later fusion. Based on this observation, we propose a new framework, named ExFuse, to bridge the gap between low-level and high-level features thus significantly improve the segmentation quality by 4.0\% in total. Furthermore, we evaluate our approach on the challenging PASCAL VOC 2012 segmentation benchmark and achieve 87.9\% mean IoU, which outperforms the previous state-of-the-art results.