Augmenting Moment Retrieval: Zero-Dependency Two-Stage Learning
Zhengxuan Wei, Jiajin Tang, Sibei Yang · 2025
Read Augmenting Moment Retrieval: Zero-Dependency Two-Stage Learning by Zhengxuan Wei, Jiajin Tang, Sibei Yang (2025) 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-2510-19622.txt.
Description
Zhengxuan Wei, Jiajin Tang, and Sibei Yang tackle moment retrieval, the task of finding the exact segment of a video that a text query describes. Existing methods suffer from scarce labeled data, fuzzy boundaries between neighboring events, and trouble telling apart similar actions like kicking and throwing. Their framework, AMR, needs no external data to fix these weaknesses.
Training runs in two stages. A cold-start stage uses curriculum learning on augmented data to build boundary and semantic awareness. A distillation stage then adapts to real data with two query sets, one frozen from the first stage and one that keeps learning, tied together by a consistency loss. Across benchmarks, AMR improved on earlier leading methods.
Catalog record
| CREATOR | Zhengxuan Wei, Jiajin Tang, Sibei Yang |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 35 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-19622