Text conditioned score tuning improves partial video retrieval accuracy
Concentrate After Imagination: Text-Conditioned Evidence Grounding for Partially Relevant Video Retrieval
Computer Vision and Pattern Recognition
Summary
Sometimes when searching videos for short moments described by text, a video might look relevant just because a small part is similar, even if most is unrelated. The authors call this problem a concentration bottleneck. They propose a method named TRACE that carefully checks evidence in videos based on the query, refining scores to better find truly relevant moments. On several video benchmarks, TRACE improved the ability to retrieve correctly matched video parts more than previous approaches.
What this means in practice
- •For video search engine developers: Enable search engines to find short relevant moments more accurately in long videos by refining scores with query-aware evidence.
- •For digital content management teams: Improve video content tagging and indexing by better detecting partial matches between text queries and video segments.
Authors
Shuaiqi Cheng, Siyu You, Yanbi Wu, Yuxi Chen, Jiahao Zhang, Xuming Hu
Abstract
Partially Relevant Video Retrieval (PRVR) retrieves untrimmed videos when queries describe only short moments. Although recent methods improve local representations, uncertainty modeling, and global context, final ranking often still trusts the strongest local response; a coincidentally similar fragment can therefore produce an unsupported peak. We identify this failure as the query-agnostic concentration bottleneck and propose TRACE, a score-level evidence verification operator for PRVR. Given a query and global video registers, TRACE activates query-relevant registers, routes their support to frame-level evidence, and smoothly marginalizes alternative query-to-register-to-frame paths before localized temporal selection. Unlike representation-level feature fusion, TRACE uses this evidence only as a query-conditioned residual calibration of the original local score. On ActivityNet Captions, Charades-STA, and TVR, TRACE achieves the best SumR on all three benchmarks and improves the DreamPRVR backbone by 1.2, 1.1, and 1.5 points, respectively. Ablation, routing-corruption, hard-negative, and cross-backbone transfer analyses support the interpretation that the gains arise from query-conditioned evidence verification rather than a generic score offset.