Research map / Video / Video-language understanding
Learning Transferable Temporal Primitives for Video Reasoning via Synthetic Videos
Songtao Jiang, Sibo Song, Chenyi Zhou, Yuan Wang, Ruizhe Chen, Tongkun Guan, Ruilin Luo, Yan Zhang, Zhihang Tang, Yuchong Sun, Hang Zhang, Zhibo Yang and 3 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 |
|---|---|---|---|
| post-train with SynRL (synthetic SFT + GRPO) on Qwen3-VL-4B | Qwen3-VL-4B · TOMATO | accuracy: 32.1 -> 36.7 | Improved |
| post-train with SynRL on Qwen3-VL-8B | Qwen3-VL-8B · TOMATO | accuracy: 33.2 -> 38.1 | Improved |
| post-train with SynRL | Qwen3-VL-4B · RexTime | R@0.3: 26.2 -> 38.8 | Improved |
| post-train with SynRL | Qwen3-VL-4B · Charades-STA | mIoU: 41.9 -> 47.0 | Improved |
| prompt Qwen3-VL-4B with CoT only, no SynRL training | Qwen3-VL-4B · Charades-STA | mIoU: 41.9 -> 34.5 | Hurt |
| add cold-start SFT before RL (ablation) | SynRL on Qwen3-VL-4B · TOMATO | accuracy: 32.8 -> 36.9 | Improved |
| add RL (GRPO) after SFT (ablation) | SynRL on Qwen3-VL-4B · TOMATO | accuracy: 34.2 -> 36.9 | Improved |
| use 7.7K synthetic CoT cold-start instead of Video-R1 165K CoT (ablation) | Qwen3-VL-4B · TOMATO | accuracy: 28.2 -> 34.2 | Improved |
| cold-start with Video-R1 CoT-165K (ablation) | Qwen3-VL-4B · vinoground | accuracy: 40.8 -> 36.6 | Hurt |
| add second RL round with 3K longer, harder long-term videos (ablation) | SynRL on Qwen3-VL-4B · TOMATO | accuracy: 36.7 -> 37.0 | Improved |
| add second RL round (ablation) | SynRL on Qwen3-VL-4B · Video-MME | accuracy: 62.0 -> 62.4 | Improved |
| post-train with SynRL | Qwen3-VL-4B · Minerva | accuracy: 32.5 -> 32.8 | No clear effect |
Closest papers by meaning
- Before It Fades: Reinforcing Temporal Representations at Inference Time in VideoLLMs (NeurIPS 2026 · similarity 0.90)
- TOMATO: Assessing Visual Temporal Reasoning Capabilities in Multimodal Foundation Models (ICLR 2025 · similarity 0.89)
- MotionBench: Benchmarking and Improving Fine-grained Video Motion Understanding for Vision Language Models (CVPR 2025 · similarity 0.89)
- VideoHallu: Evaluating and Mitigating Multi-modal Hallucinations on Synthetic Video Understanding (NeurIPS 2025 · similarity 0.88)
- Time-R1: Post-Training Large Vision Language Model for Temporal Video Grounding (NeurIPS 2025 · similarity 0.88)
- EventLens: Event-Structure Reinforcement Learning for Video Understanding (NeurIPS 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.
