Reproduce “VerdictAD: Few-Shot Anomaly Detection via Decision-Guided 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
Shangquan Sun, Wenjun Luo, Yuanzhi Wang, Ruiqi Li, Ruoyu Chen, Jiachen Liu, Shijian Lu
VerdictAD refines few-shot industrial anomaly detectors with a frozen decision model that checks scores and local maps. Global judgments adjust image scores, while crop- and box-based judgments reweight map responses; the detector is not trained. On four-shot MVTec LOCO AD, SubspaceAD image AUROC rises from 67.5 ± 1.7% to 75.3 ± 2.0%. Refinement cannot restore evidence absent from the original map.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.