Reproduce “Continual-MEGA: A Large-scale Benchmark for Generalizable Continual 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
Geonu Lee, Yujeong Oh, Geonhui Jang, Soyoung Lee, Jeonghyo Song, Sungmin Cha, YoungJoon Yoo
Continual-MEGA introduces a benchmark for adapting defect detectors as product categories arrive and testing them on categories withheld from adaptation. The authors propose ADCT, a CLIP-based baseline that uses task-specific adapters and synthetic anomaly features. In Scenario 1, ADCT reaches Image-AUROC 78.9, Pixel-AP 32.7, and mean ACC 55.8; this result applies to that setting, not every metric or scenario.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.