Reproduce “Fault-Guided Cross-Layer Prediction for Unsupervised 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
Tao Wang
The authors present a single-class unsupervised anomaly detector that predicts representations between blocks of a frozen vision transformer. They select source and target blocks using drops in adjacent-layer cosine similarity, then use prediction residuals as anomaly signals. On Real-IAD at 448, image-level AUROC is 94.27%, versus Dinomaly2’s 91.98%; this advantage does not extend to pixel metrics or all benchmarks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.