The authors built a unified benchmark around two challenge tracks.
Hidden tests expose persistent gaps in real-world visual defect detection
Across manufacturing and retail benchmarks, specialized models led, but rare defects, distribution shifts and computational efficiency remained unresolved.
Big Tech
Lars Heckler-Kram · Dorian Henning · Ashwin Vaidya · Jan-Hendrik Neudeck · Ulla Scheler · Anton Milan · +3 more
MVTec Software GmbH · Technical University of Munich · Amazon · Intel · NVIDIA
Research Digest··3 min read
0 Challenge evaluated visual anomaly-detection systems on hidden test data from industrial manufacturing and retail logistics.
Why this paper
From Amazon and 4 others
In one line
Supervised anomaly detection saturates on common defects but fails on rare ones; zero-shot VLMs are inferior; efficiency is neglected.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors
- ✓Reports numbers on named benchmarks (2 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§