Reproduce “TED:Text-Axis Evidence Decomposition for Prompted Anomaly Localization” 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

JinYoung Kim, Geonho Kim, GiJeong Park, Geonu Lee, YoungJoon Yoo

The authors introduce TED, a post-hoc method for local anomaly maps. TED compares suspicious patches’ support from source defect examples with support from source normal patches the host mistakes for anomalies. It keeps the host and prompts fixed and uses no target-domain calibration. Pixel-level localization improves in most studied adapted-host settings, with effects varying by host and metric.

Summary by alphaXiv.

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