Reproduce “MESAM: Enhancing SAM2 for Few-Shot Segmentation through Semantic-Visual Memory Generation and Refinement” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Tianyu Zou, Shengwu Xiong, Rui Wu, Ruilin Yao, Yi Rong

MESAM uses category-level cues to guide SAM2’s memory in few-shot segmentation, where one labeled image helps segment another image of the same category. It combines CLIP-derived semantic priors with visual features and calibrates attention to suppress background matches. On COCO-20ᵢ 1-shot evaluation with SAM2-S, MESAM reaches 67.3% mean mIoU versus 62.3% for FSSAM.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.