When SAM Meets Sonar Images

TL;DR

This paper explores the application of the Segment Anything Model (SAM) to sonar images, evaluating its performance and enhancing it through fine-tuning techniques to improve segmentation accuracy in this domain.

Contribution

It is the first comprehensive investigation of SAM's performance on sonar images, including evaluation and fine-tuning methods to adapt SAM for sonar segmentation tasks.

Findings

Fine-tuning significantly improves SAM's performance on sonar images.

SAM can be effectively adapted to non-natural image domains with proper fine-tuning.

The study provides a foundation for applying SAM in sonar imaging applications.

Abstract

Segment Anything Model (SAM) has revolutionized the way of segmentation. However, SAM's performance may decline when applied to tasks involving domains that differ from natural images. Nonetheless, by employing fine-tuning techniques, SAM exhibits promising capabilities in specific domains, such as medicine and planetary science. Notably, there is a lack of research on the application of SAM to sonar imaging. In this paper, we aim to address this gap by conducting a comprehensive investigation of SAM's performance on sonar images. Specifically, we evaluate SAM using various settings on sonar images. Additionally, we fine-tune SAM using effective methods both with prompts and for semantic segmentation, thereby expanding its applicability to tasks requiring automated segmentation. Experimental results demonstrate a significant improvement in the performance of the fine-tuned SAM.