Improving Temporal Action Segmentation via Constraint-Aware Decoding
Yeo Keat Ee, Debaditya Roy, Chen Li, Hao Zhang, Basura Fernando · 2026
Read Improving Temporal Action Segmentation via Constraint-Aware Decoding by Yeo Keat Ee, Debaditya Roy, Chen Li, Hao Zhang, Basura Fernando (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/arxiv-2605-10149.txt.
Description
Yeo and four co-authors improve temporal action segmentation, the task of splitting raw videos into tagged action segments. Fully supervised methods have grown, but action spread, fuzzy edges and costly tagging persist, above all in fresh or low-resource domains. Grammar-based methods add structural priors yet depend on complex parsing that limits scalability. The authors propose a lightweight, constraint-based polish system instead.
Stats shape priors, including shift trust, action edge sets and per-class length, are drawn straight from tagged data and folded into a tweaked Viterbi reading method. That allows run-time polish with no retrain and no added model bulk. The approach lifts both fully and semi-trained splitting models by fixing shape guess errors while staying lean, with code released publicly.
Catalog record
| CREATOR | Yeo Keat Ee, Debaditya Roy, Chen Li, Hao Zhang, Basura Fernando |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 33 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.10149