Structure-aware anomaly detection via semantic prototype reconstruction in hyperbolic space
Yanjun Feng, Jun Liu, Yonggang Gai · 2026
Read Structure-aware anomaly detection via semantic prototype reconstruction in hyperbolic space by Yanjun Feng, Jun Liu, Yonggang Gai (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-13493839.txt.
Description
Feng, Liu, and Gai propose a new way to spot defects in factory photographs, such as scratches, cracks, and missing parts, by working in hyperbolic space. That is a kind of curved geometry that expands rapidly outward. Their 2026 framework learns typical patterns drawn from flawless products and embeds them in that curved space. It then tries to reconstruct each new image guided by the patterns. Where reconstruction fails badly, the system marks a defect and pinpoints its location.
On the widely used MVTec AD benchmark, the method scored near the ceiling on the standard AUROC measure at both whole-image and single-pixel levels. That measure grades how well a detector separates faulty from clean items. The authors say it beat current leading methods across several industrial datasets. They argue the curved space suits defect data because normal products form a hierarchy of variations that flat geometry squeezes together.
Catalog record
| CREATOR | Yanjun Feng, Jun Liu, Yonggang Gai |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 41 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |