Cell image segmentation by Feature Random Enhancement Module

TL;DR

This paper introduces the Feature Random Enhancement Module, a training-only technique that improves deep neural network feature extraction for cell image segmentation by emphasizing far-layer features, leading to higher accuracy without extra test costs.

Contribution

The paper proposes a novel Feature Random Enhancement Module that enhances features in training to improve segmentation accuracy, especially for far layers from the loss function.

Findings

Improved segmentation accuracy on cell image datasets

No increase in computational cost during testing

Effective enhancement of far-layer features during training

Abstract

It is important to extract good features using an encoder to realize semantic segmentation with high accuracy. Although loss function is optimized in training deep neural network, far layers from the layers for computing loss function are difficult to train. Skip connection is effective for this problem but there are still far layers from the loss function. In this paper, we propose the Feature Random Enhancement Module which enhances the features randomly in only training. By emphasizing the features at far layers from loss function, we can train those layers well and the accuracy was improved. In experiments, we evaluated the proposed module on two kinds of cell image datasets, and our module improved the segmentation accuracy without increasing computational cost in test phase.