# Reason-Then-Retrieve for CoVR-R with Structured Edit Prompts and Dense-Sparse Fusion

arXiv: https://arxiv.org/abs/2606.02450 · HTML: https://mlsanity.com/paper/2606.02450/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 7 of its 16 references that are in the map.
- First posted on arXiv 1 Jun 2026.

## Builds on (most cited first, 7 of 7)

- [Weakly-Supervised Temporal Action Localization by Inferring Salient Snippet-Feature](https://mlsanity.com/paper/2303.12332/) (2023)
- [Action Quality Assessment via Hierarchical Pose-guided Multi-stage Contrastive Regression](https://mlsanity.com/paper/2501.03674/) (2025)
- [Robust Disentangled Counterfactual Learning for Physical Audiovisual Commonsense Reasoning](https://mlsanity.com/paper/2502.12425/) (2025)
- [DC-SAM: In-Context Segment Anything in Images and Videos via Dual Consistency](https://mlsanity.com/paper/2504.12080/) (2025)
- [Explainable Action Form Assessment by Exploiting Multimodal Chain-of-Thoughts Reasoning](https://mlsanity.com/paper/2512.15153/) (2025)
- [Towards Balanced Multi-Modal Learning in 3D Human Pose Estimation](https://mlsanity.com/paper/2501.05264/) (2025)
- [CoVR-R:Reason-Aware Composed Video Retrieval](https://mlsanity.com/paper/2603.20190/) (2026)

## Built on by (most cited first, 1 of 1)

- [Beyond Similarity: Foundation Models as an Efficient Backbone for Training-Free Composed Video Retrieval](https://mlsanity.com/paper/2609.10008/) (2026)

## Abstract (arXiv)

CoVR-R studies reason-aware composed video retrieval: given a reference video and an edit instruction, the system must retrieve the target video that satisfies the edit. The main difficulty is that the target is not described directly; it must be inferred from fine-grained changes in object identity, action order, final state, hand interaction, and scene transition. We build a zero-shot reason-then-retrieve pipeline around Qwen3.5-27B. For each gallery video, the model generates a retrieval-oriented structured description and a dense embedding by pooling generated-token hidden states with token-dependent weights. For each query, the model first performs edit reasoning over the reference video and instruction, then generates a target-video description whose hidden states serve as the query embedding. We complement dense retrieval with a TF-IDF branch over the generated texts and fuse the two rankings with split-specific weights. On validation, the current best submission reaches 80.81 at R@1, 94.86 at R@5, 97.11 at R@10, and 98.59 at R@50. On the blind test split, it reaches 89.73 at R@1, 95.79 at R@5, 96.63 at R@10, and 97.98 at R@50.