Research map / Video / Video-language understanding
Open-o3-Video: Grounded Video Reasoning with Explicit Spatio-Temporal Evidence
Jiahao Meng, Xiangtai Li, Haochen Wang, Tan Yue, Tao Zhang, Lingdong Kong, Yunhai Tong, Anran Wang, Zhiyang Teng, Yujing Wang, Zhuochen 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 RL stage after SFT (ablation) | Open-o3-Video · V-STAR | mAM: 28.5 -> 33.7 | Improved |
| add SFT cold start before RL (ablation) | Open-o3-Video · V-STAR | mAM: 30.4 -> 33.7 | Improved |
| use GSPO instead of GRPO (ablation) | SFT+RL · V-STAR | mAM: 32.8 -> 33.7 | Improved |
| add adaptive temporal proximity (ablation) | Open-o3-Video reward · V-STAR | mAM: 33.0 -> 33.7 | Improved |
| add temporal gating (ablation) | Open-o3-Video reward · V-STAR | mAM: 32.3 -> 33.7 | Improved |
| add spatio-temporal grounded training data (ablation) | Open-o3-Video · V-STAR | mAM: 28.3 -> 33.7 | Improved |
| add own annotated data on top of VideoEspresso (ablation) | Open-o3-Video · V-STAR | mAM: 31.1 -> 33.7 | Improved |
| adaptive sigma instead of fixed sigma=1.0 (ablation) | Open-o3-Video · V-STAR | mAM: 32.6 -> 33.7 | Improved |
| add 30k instead of 15k Video-R1 QA data (ablation) | Open-o3-Video · V-STAR | mAM: 33.7 -> 31.7 | Hurt |
| add spatio-temporal evidence (vs text-only reasoning) (ablation) | Open-o3-Video · V-STAR mLGM | mLGM: 41.0 -> 46.6 | Improved |
| remove predicted evidence frames vs random removal (ablation) | Open-o3-Video · VideoMME subset (150 questions) | accuracy: 66.0 vs 68.0 | Improved |
| confidence-aware voting instead of majority voting | Open-o3-Video · WorldSense | accuracy: 37.3 -> 38.5 | Improved |
Later papers that compared with this method
2 later accepted papers report Open-o3-video-7B as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- VisionCoach: Reinforcing Grounded Video Reasoning via Visual-Perception Prompting (ECCV 2026 · better on 3 of 3 shared results)
- MotionChain: Fine-Grained Video Motion Understanding via Structured Decomposition (ECCV 2026 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Sa2VA-8B — Sa2VA: Marrying SAM2 With MLLM for Dense Grounded Understanding of Images and Videos (TPAMI 2026) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Video Spatial Reasoning with Object-Centric 3D Rollout (AAAI 2026 · similarity 0.90)
- Motion-Grounded Video Reasoning: Understanding and Perceiving Motion at Pixel Level (CVPR 2025 · similarity 0.90)
- Look, Remember and Reason: Grounded Reasoning in Videos with Language Models (ICLR 2024 · similarity 0.89)
- OpenCoF: Learning to Reason Through Video Generation (NeurIPS 2026 · similarity 0.89)
- SciVideoBench: Benchmarking Scientific Video Reasoning in Large Multimodal Models (ICML 2026 · similarity 0.89)
- Traceable Evidence Enhanced Visual Grounded Reasoning: Evaluation and Method (ICLR 2026 · 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.
