Research map / Video / Video-language understanding
VideoTrace-R1: Long Video-based Retrieval-Augmented Generation via Reinforcement Learning
Zongsheng Cao, Anran Liu, Jun Xie, Feng Chen, Lang Chen, Jing Li, Zhepeng Wang, Zigan Wang
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 trace-grounded GRPO after SFT (ablation) | VideoTrace-R1-SFT-7B · Video-Holmes | accuracy: 30.87 -> 43.51 | Improved |
| add trace-grounded GRPO after SFT (ablation) | VideoTrace-R1-SFT-7B · CG-Bench-Reasoning | accuracy: 23.28 -> 31.10 | Improved |
| add trace-grounded GRPO after SFT (ablation) | VideoTrace-R1-SFT-7B · VRBench | accuracy: 60.97 -> 76.23 | Improved |
| add TRT index to the reward (ablation) | VideoTrace · MLVU (ablation) | accuracy: 69.7 -> 72.5 | Improved |
| add TRT index to the reward (ablation) | VideoTrace · VideoMME (ablation) | accuracy: 72.3 -> 74.7 | Improved |
| add TRT index to the reward (ablation) | VideoTrace · LongVideoBench (ablation) | accuracy: 62.7 -> 66.1 | Improved |
| add GRPO stage on top of SFT (ablation) | VideoTrace · MLVU (ablation) | accuracy: 71.2 -> 72.5 | Improved |
| add GRPO stage on top of SFT (ablation) | VideoTrace · VideoMME (ablation) | accuracy: 73.5 -> 74.7 | Improved |
| add GRPO stage on top of SFT (ablation) | VideoTrace · LongVideoBench (ablation) | accuracy: 65.0 -> 66.1 | Improved |
| append TRT event chains to prompt of frozen host | Qwen2.5-VL 7B · MLVU (frozen host, 7B) | accuracy: 68.8 -> 71.6 | Improved |
| append TRT event chains to prompt of frozen host | Qwen2.5-VL 3B · MLVU | accuracy: 66.2 -> 69.9 | Improved |
| append TRT event chains to prompt of frozen host | LongVU 7B · MLVU | accuracy: 65.4 -> 70.1 | Improved |
Closest papers by meaning
- Find, Fix, Reason: Context Repair for Video Reasoning (ICML 2026 · similarity 0.89)
- When Thinking Drifts: Evidential Grounding for Robust Video Reasoning (NeurIPS 2025 · similarity 0.89)
- Reason in the Words You Speak: Idiolectal Paraphrasing Off-Policy Traces for Reasoning Distillation in VideoLLMs (NeurIPS 2026 · similarity 0.89)
- ReWatch-R1: Boosting Complex Video Reasoning in Large Vision-Language Models through Agentic Data Synthesis (ICLR 2026 · similarity 0.88)
- Before It Fades: Reinforcing Temporal Representations at Inference Time in VideoLLMs (NeurIPS 2026 · similarity 0.88)
- Video-R1: Reinforcing Video Reasoning in MLLMs (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.
