R3-CoVR achieves 91.9% R@1 and 98.2% R@10 on zero-shot CoVR by reasoning about edit after-effects with Qwen3-VL-8B, retrieving with SigLIP-2, and constraint-aware re-ranking.
arXiv:2502.20826, 2025
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Reason, Retrieve, Re-rank: A Zero-Shot Reasoning-Aware Framework for Composed Video Retrieval
R3-CoVR achieves 91.9% R@1 and 98.2% R@10 on zero-shot CoVR by reasoning about edit after-effects with Qwen3-VL-8B, retrieving with SigLIP-2, and constraint-aware re-ranking.