TriPath3DNet: an efficient real-time model for multi-class classification in real-life surveillance videos of fixed duration
K Mohanarangan, P Palanisamy · 2026
Read TriPath3DNet: an efficient real-time model for multi-class classification in real-life surveillance videos of fixed duration by K Mohanarangan, P Palanisamy (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/PMC13547256.txt.
Description
K Mohanarangan and P Palanisamy presented in 2026 a video analysis network built for the messy reality of surveillance footage. That means brief motion-triggered clips with blocked views, shifting light, bad weather, and quiet anomalies like loitering. Their model reads each clip through three time lenses at once, catching quick movements, longer context, and changes between frames, all on a compact backbone suited to live use.
Across four datasets, including a well-known crime video benchmark, the model reached accuracy up to about 95 percent while judging a 50-frame clip in roughly 130 milliseconds. It beat several larger transformer systems on the anomaly-heavy tests, and removing any of its three branches hurt performance. Attention maps show the model looking at the right places at the right moments.
Catalog record
| CREATOR | K Mohanarangan, P Palanisamy |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 68356 |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |