Reason, Retrieve, Re-rank: A Zero-Shot Reasoning-Aware Framework for Composed Video Retrieval
Ali Alavi · 2026
Read Reason, Retrieve, Re-rank: A Zero-Shot Reasoning-Aware Framework for Composed Video Retrieval by Ali Alavi (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-2606-00910.txt.
Description
Alavi builds a zero-shot pipeline for finding an edited video in this 2026 paper. Composed video retrieval starts from a reference clip plus a text edit. The goal is the clip that the edit would produce. His entry targets a reason-aware challenge at a CVPR 2026 workshop, with no training allowed.
His pipeline, R3-CoVR, runs in three steps, all from frozen models. A multimodal model first reasons about the edit's after-effects: state changes, action phases, scene, camera, and tempo. It writes a short post-edit description. A contrastive encoder then searches the gallery with that text. Last, the same multimodal model judges the shortlist against the planned result and re-ranks it. On the test set, the pipeline hits 91.9 percent recall at rank one, they report in challenge terms. The re-ranker alone lifts that score from 72.7 to 91.9.
Catalog record
| CREATOR | Ali Alavi |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 15 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.00910