Mix-up Self-Supervised Learning for Contrast-agnostic Applications

TL;DR

This paper introduces a novel mix-up self-supervised learning framework tailored for contrast-agnostic applications like medical imaging, improving representation learning where traditional contrastive methods struggle due to high image similarity.

Contribution

It proposes the first mix-up self-supervised learning approach for contrast-agnostic tasks, combining image reconstruction and transparency prediction to enhance feature variance.

Findings

Achieved 2.5% to 7.4% improvement in top-1 accuracy on benchmark datasets.

Validated effectiveness across two benchmark datasets.

Addresses the challenge of low variance in contrast-agnostic image representations.

Abstract

Contrastive self-supervised learning has attracted significant research attention recently. It learns effective visual representations from unlabeled data by embedding augmented views of the same image close to each other while pushing away embeddings of different images. Despite its great success on ImageNet classification, COCO object detection, etc., its performance degrades on contrast-agnostic applications, e.g., medical image classification, where all images are visually similar to each other. This creates difficulties in optimizing the embedding space as the distance between images is rather small. To solve this issue, we present the first mix-up self-supervised learning framework for contrast-agnostic applications. We address the low variance across images based on cross-domain mix-up and build the pretext task based on two synergistic objectives: image reconstruction and transparency prediction. Experimental results on two benchmark datasets validate the effectiveness of our method, where an improvement of 2.5% ~ 7.4% in top-1 accuracy was obtained compared to existing self-supervised learning methods.