Finding Optimal Video Moment without Training: Gaussian Boundary Optimization for Weakly Supervised Video Grounding
Sunoh Kim, Kimin Yun, Daeho Um · 2026
Read Finding Optimal Video Moment without Training: Gaussian Boundary Optimization for Weakly Supervised Video Grounding by Sunoh Kim, Kimin Yun, Daeho Um (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-2602-03071.txt.
Description
Finding the right moment in a video without training is the aim of Sunoh Kim and 2 co-authors in a 2026 paper. Weakly supervised grounding locates a queried segment using only video-sentence pairs, without exact boundary labels. Recent methods represent candidate segments with Gaussian proposals, bell-shaped ranges over time, but convert them to boundaries with heuristic rules that localize poorly in practice.
Their framework, Gaussian Boundary Optimization, instead solves a principled optimization problem. It balances covering the proposal against keeping the segment compact. A closed-form solution is derived, with optimality analyzed under different penalty regimes. The method is training-free and fits single or mixture proposal systems. Experiments across standard benchmarks reached state-of-the-art localization, with efficiency and generalizability shown.
Catalog record
| CREATOR | Sunoh Kim, Kimin Yun, Daeho Um |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 61 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-03071