Reproduce “PTQ4SAM3: Semantics-Preserving Post-Training Quantization for Segment Anything Model 3” 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

Xi Xiao, Yunbei Zhang, Lin Zhao, Janet Wang, Xinru Jiang, Yanshu Li, Chenrui Ma, Pan Wang and 2 more

SPQ quantizes SAM 3 while targeting prompt-alignment drift in image and video segmentation. It weights calibration by text-concept scores, protects low-variance residual features, and corrects mean error at video-memory interfaces; in W4A8, RefCOCO [email protected] reaches 72.5 versus 42.1 for PTQ4SAM, while COCO mask AP is 54.8 versus 54.1, with several components kept at higher precision.

Summary by alphaXiv.

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