Reproduce “Spatiotemporal Dynamics-Aware Industrial Anomaly Detection and Reasoning” 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
S-Wei Chen, Siangyu Yang, Yi-Chen Jan, Chun-Yi Lee, Shao-Yuan Lo
MUSTARD uses optical flow to select informative frames and combines raw images, local motion trajectories, and a global motion curve. It compares these descriptions with patterns from normal videos to score anomalies and generate explanations without updating model weights. On Phys-AD, its video-level AUROC averages 68.1% across 22 categories, versus 66.9% for MNAD.r; it still depends on the underlying MLLM’s perception.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.