Reproduce “Efficient SAM3 Adaptation for Multi-Class Semantic Segmentation via Dense Competitive Representations” 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
Wenbin Liao, Hao Zhu, Yike Ma, Hao Jiang, Feng Dai
DCR-SAM adapts SAM3 for dense multi-class semantic segmentation by sharing image features across classes and refining class text with image context. A background token and joint competition produce class scores at each pixel. On ADE20K validation, it reports 68.7 mIoU versus 60.3 for InternImage-H. At 150 classes, it reports 244 ms versus 8,038 ms for native SAM3, under the reported setup. Results use supervised training on each dataset.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.