Research map / Video / Video-language understanding
VideoSearch-R1: Iterative Video Retrieval and Reasoning via Soft Query Refinement
Seohyun Lee, Seoung Choi, Dohwan Ko, Jongha Kim, Hyunwoo Kim
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Stage 2 GRPO after SFT (ablation) | VideoSearch-R1 · DiDeMo-FIG | VCMR 0.3/R@1: 20.4 -> 33.3 | Improved |
| add Stage 2 GRPO after SFT (ablation) | VideoSearch-R1 · DiDeMo-FIG | VCMR 0.7/R@1: 14.0 -> 19.7 | Improved |
| add SFT Stage 1 to zero-shot model (ablation) | Qwen3-VL-2B · DiDeMo-FIG | VR R@1: 54.8 -> 57.4 | Improved |
| add SFT Stage 1 to zero-shot model (ablation) | Qwen3-VL-2B · DiDeMo-FIG | VCMR 0.3/R@1: 22.0 -> 20.4 | No clear effect |
| add retrieval reward R_ret in RL (ablation) | VideoSearch-R1 · DiDeMo-FIG | VR R@1: 57.4 -> 58.0 | Improved |
| add retrieval reward R_ret in RL (ablation) | VideoSearch-R1 · DiDeMo-FIG | VCMR 0.3/R@1: 20.4 -> 18.6 | Hurt |
| add verification reward R_verif (ablation) | VideoSearch-R1 · DiDeMo-FIG | VER accuracy: 65.0 -> 75.0 | Improved |
| add temporal grounding reward R_time (ablation) | VideoSearch-R1 · DiDeMo-FIG | VCMR 0.3/R@1: 19.3 -> 33.3 | Improved |
| add temporal grounding reward R_time (ablation) | VideoSearch-R1 · DiDeMo-FIG | VR R@1: 59.7 -> 59.0 | Mixed |
| use soft query refinement instead of hard query refinement (ablation) | VideoSearch-R1 · ActivityNet-FIG | VR R@1: 57.6 -> 61.1 | Improved |
| use soft query refinement instead of hard query refinement (ablation) | VideoSearch-R1 · ActivityNet-FIG | generated tokens: 26.8 -> 8.0 | Improved |
| add soft query refinement and RL on top of fine-tuned baseline | Qwen3-VL-2B (FT) · DiDeMo-FIG | VCMR 0.3/R@1: 23.6 -> 33.3 | Improved |
Closest papers by meaning
- Vgent: Graph-based Retrieval-Reasoning-Augmented Generation For Long Video Understanding (NeurIPS 2025 · similarity 0.90)
- AdaVideoRAG: Omni-Contextual Adaptive Retrieval-Augmented Efficient Long Video Understanding (NeurIPS 2025 · similarity 0.88)
- Text Proxy: Decomposing Retrieval from a 1-to-N Relationship into N 1-to-1 Relationships for Text-Video Retrieval (AAAI 2025 · similarity 0.88)
- Video-RAG: Visually-aligned Retrieval-Augmented Long Video Comprehension (NeurIPS 2025 · similarity 0.88)
- VideoEspresso: A Large-Scale Chain-of-Thought Dataset for Fine-Grained Video Reasoning via Core Frame Selection (CVPR 2025 · similarity 0.88)
- Mitigating Semantic Collapse in Partially Relevant Video Retrieval (NeurIPS 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
