Research map / Video / Video-language understanding
VideoAuto-R1: Video Auto Reasoning via Thinking Once, Answering Twice
Shuming Liu, Mingchen Zhuge, Changsheng Zhao, Jun Chen, Lemeng Wu, Zechun Liu, Chenchen Zhu, Zhipeng Cai, Chong Zhou, Haozhe Liu, Ernie Chang, Saksham Suri and 11 more
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 |
|---|---|---|---|
| Use CoT instead of direct inference in Video-R1/Time-R1/VideoChat-R1 | RL video reasoning models · VideoMME | accuracy: 64.6 -> 64.3 (Video-R1); 65.9 -> 63.8 (Time-R1) | Hurt |
| Use CoT instead of direct inference | VideoChat-R1 · VideoMMMU | accuracy: 52.3 -> 55.7 | Improved |
| Use CoT instead of direct inference | Video-R1 · Charades-STA | mIoU: 42.0 -> 34.9 | Hurt |
| Replace RL-with-thinking by answer-think-answer with early exit (ablation) | Qwen2.5-VL-7B · VideoMMMU | accuracy: 56.4 -> 58.6 | Improved |
| Replace RL-with-thinking by answer-think-answer with early exit (ablation) | Qwen2.5-VL-7B · avg response length | tokens: 149 -> 44 | Improved |
| Use inference-based early exit instead of training-based think/no-think selection (ablation) | Qwen2.5-VL-7B · VideoMMMU | accuracy: 55.7 -> 58.6 | Improved |
| Training-based auto-thinking vs no-think (ablation) | Qwen2.5-VL-7B · MVBench | accuracy: 71.1 -> 70.5 | Hurt |
| Early exit (auto) vs always using 2nd answer (ablation) | VideoAuto-R1 · VideoMMMU | accuracy: 58.7 -> 58.6 | No clear effect |
| Weight final answer higher (0.9:1.1 vs 1:1) (ablation) | VideoAuto-R1 · VideoMMMU | accuracy: 56.1 -> 56.4 | Improved |
| Add fallback reward (ablation) | VideoAuto-R1 · VideoMMMU | accuracy: 56.4 -> 58.6 | Improved |
| Add fallback reward at 0.9:1.1 weights (ablation) | VideoAuto-R1 · VideoMME | accuracy: 66.0 -> 67.3 | Improved |
| Dual-answer RL vs SFT direct answer (ablation) | Qwen2.5-VL-7B · Charades-STA | mIoU: 56.3 -> 60.0 | Improved |
Later papers that compared with this method
2 later accepted papers report Video-Auto-R1 as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Video-FLAIR: Not Whether to Reason, But How (NeurIPS 2026 · better on 3 of 4 shared results)
- VideoKR: Towards Knowledge- and Reasoning-Intensive Video Understanding (ICML 2026 · better on 2 of 3 shared results)
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.
- Time-R1 — Time-R1: Post-Training Large Vision Language Model for Temporal Video Grounding (NeurIPS 2025) · 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
- Enhancing Video-LLM Reasoning via Agent-of-Thoughts Distillation (CVPR 2025 · similarity 0.92)
- VideoReasonBench: Can MLLMs Perform Vision-Centric Complex Video Reasoning? (ICLR 2026 · similarity 0.91)
- VideoEspresso: A Large-Scale Chain-of-Thought Dataset for Fine-Grained Video Reasoning via Core Frame Selection (CVPR 2025 · similarity 0.91)
- Video-R1: Reinforcing Video Reasoning in MLLMs (NeurIPS 2025 · similarity 0.90)
- When Thinking Drifts: Evidential Grounding for Robust Video Reasoning (NeurIPS 2025 · similarity 0.90)
- Divide and Conquer: Exploring Language-centric Tree Reasoning for Video Question-Answering (ICML 2025 · similarity 0.90)
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.
