Reproduce “Contextual Affinity Distillation for Image 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
Jie Zhang, Masanori Suganuma, Takayuki Okatani
DSKD detects structural defects and logical errors in industrial inspection by pairing a local student, which reproduces teacher features, with a global student trained to represent image-wide relationships. On MVTec LOCO AD, it reports a mean image-level AUROC of 0.840, versus 0.833 for GCAD, and localization sPRO of 0.730 versus 0.701. GCAD scores higher on logical-anomaly AUROC: 0.860 versus 0.812.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.