Localized Feature Aggregation Module for Semantic Segmentation

TL;DR

This paper introduces a Localized Feature Aggregation Module that enhances semantic segmentation by efficiently combining encoder and decoder features, improving accuracy while reducing computational costs.

Contribution

The paper presents a novel aggregation module that better recovers positional information and reduces computation compared to traditional concatenation and attention methods.

Findings

Outperforms conventional methods on Drosophila cell images

Achieves higher segmentation accuracy on COVID-19 datasets

Reduces computational cost through localized attention

Abstract

We propose a new information aggregation method which called Localized Feature Aggregation Module based on the similarity between the feature maps of an encoder and a decoder. The proposed method recovers positional information by emphasizing the similarity between decoder's feature maps with superior semantic information and encoder's feature maps with superior positional information. The proposed method can learn positional information more efficiently than conventional concatenation in the U-net and attention U-net. Additionally, the proposed method also uses localized attention range to reduce the computational cost. Two innovations contributed to improve the segmentation accuracy with lower computational cost. By experiments on the Drosophila cell image dataset and COVID-19 image dataset, we confirmed that our method outperformed conventional methods.