Reproduce “Sam-Correction: Fully Adaptive Label Noise Reduction for Medical Image Segmentation” 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

Takuro Shimaya, Masahiro Saiko

SAM-C uses Segment Anything (SAM) to revise noisy segmentation labels, then trains a segmentation network on the corrected masks. It prompts SAM with the network’s predictions and repeats correction and training. Across four conditions, its models had higher test-set DSC than Spatial Correction (SC), while the clean-label oracle remained higher. In LIDC-IDRI’s “worst” condition, SC’s corrected labels had higher DSC against clean references than SAM-C’s.

Summary by alphaXiv.

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