Reproduce “Knowledge-Guided Distributional Prompt Learning with Visual Hierarchy for Zero-Shot Anomaly Detection” 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
Jiyul Ham, Seongho Yoon, Yonggon Jung, Sangjae Lee, Jun-Geol Baek
HiDist addresses zero-shot anomaly detection by modeling anomalous prompts as learned deviations from corresponding normal prompts. It assigns token-wise distributions to prompts and uses separate global and local branches for image-level detection and pixel-level localization, aligned through shared visual features. HiDist reports 94.7 image-level AUROC on MVTec-AD versus 90.5 for Bayes-PFL, but does not lead on every localization metric.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.