PoinCLIP-VAD: a hyperbolic cross-modal fusion framework for video anomaly detection
Debi Prasad Senapati, Santosh Kumar Pani, Santos Kumar Baliarsingh, Prabhu Prasad Dev, Ramasubbareddy Somula · 2026
Read PoinCLIP-VAD: a hyperbolic cross-modal fusion framework for video anomaly detection by Debi Prasad Senapati, Santosh Kumar Pani, Santos Kumar Baliarsingh, Prabhu Prasad Dev, Ramasubbareddy Somula (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-13328543.txt.
Description
Debi Prasad Senapati and team built a hyperbolic cross-modal fusion frame for video anomaly detection. Their 2026 paper takes on a core limit of weakly watched detection: with no frame-level labels, multiple instance learning picks noisy instances and weakly ties video bits to meanings. Sight-word models help by lining up visual traits with text labels, but flat Euclidean maps struggle with fine meaning shifts and murky instance ranks under weak watching.
PoinCLIP-VAD plants visual and text traits in a shared Poincaré ball shape, where bent distance scaling airs hidden meaning links, with no set hierarchies. A two-block build pairs a scoring block for rough anomaly scores with a video-text block for fine matches, using negative Poincaré distance. On test sets, the frame hit 90.62% AUC on UCF-Crime and 86.93% average precision on XD-Violence, backing better anomaly spotting and steadier cross-modal lining-up.
Catalog record
| CREATOR | Debi Prasad Senapati, Santosh Kumar Pani, Santos Kumar Baliarsingh, Prabhu Prasad Dev, Ramasubbareddy Somula |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 79 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |