TSVC: Tripartite Learning with Semantic Variation Consistency for Robust Image-Text Retrieval
Shuai Lyu, Zijing Tian, Zhonghong Ou, Yifan Zhu, Xiao Zhang, Qiankun Ha, Haoran Luo, Meina Song · 2025
Read TSVC: Tripartite Learning with Semantic Variation Consistency for Robust Image-Text Retrieval by Shuai Lyu, Zijing Tian, Zhonghong Ou, Yifan Zhu, Xiao Zhang, Qiankun Ha, Haoran Luo, Meina Song (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-2501-10935.txt.
Description
Shuai Lyu and seven co-authors built a retrieval method in two thousand twenty-five that stays strong when image-text pairs are wrongly matched. Search systems that pair photos with captions assume the pairs are clean, but real data sets carry noisy links that drag accuracy down. Past fixes used twin models that slowly grow too alike to help.
Their method, TSVC, uses three models in distinct roles: a coordinator that hands out data, a master, and an assistant that feeds the master varied views. A soft label scheme judges how noisy each new pair is. Tests on three data sets show clear gains in accuracy even as noise rises.
Catalog record
| CREATOR | Shuai Lyu, Zijing Tian, Zhonghong Ou, Yifan Zhu, Xiao Zhang, Qiankun Ha, Haoran Luo, Meina Song |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 32 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-10935