Improving the Generalization of Segmentation Foundation Model under Distribution Shift via Weakly Supervised Adaptation

TL;DR

This paper introduces a weakly supervised self-training approach with anchor regularization and low-rank finetuning to enhance the robustness and generalization of the Segment-Anything model across diverse distribution shifts in various segmentation tasks.

Contribution

It proposes a novel self-training adaptation framework that addresses computational and pseudo-label accuracy challenges, improving SAM's performance under distribution shifts.

Findings

Outperforms pre-trained SAM on multiple downstream tasks

Enhances robustness to corrupted and medical images

Achieves better domain adaptation efficiency

Abstract

The success of large language models has inspired the computer vision community to explore image segmentation foundation model that is able to zero/few-shot generalize through prompt engineering. Segment-Anything(SAM), among others, is the state-of-the-art image segmentation foundation model demonstrating strong zero/few-shot generalization. Despite the success, recent studies reveal the weakness of SAM under strong distribution shift. In particular, SAM performs awkwardly on corrupted natural images, camouflaged images, medical images, etc. Motivated by the observations, we aim to develop a self-training based strategy to adapt SAM to target distribution. Given the unique challenges of large source dataset, high computation cost and incorrect pseudo label, we propose a weakly supervised self-training architecture with anchor regularization and low-rank finetuning to improve the robustness and computation efficiency of adaptation. We validate the effectiveness on 5 types of downstream segmentation tasks including natural clean/corrupted images, medical images, camouflaged images and robotic images. Our proposed method is task-agnostic in nature and outperforms pre-trained SAM and state-of-the-art domain adaptation methods on almost all downstream tasks with the same testing prompt inputs.