Research map / Video / Video-language understanding
TempR1: Improving Temporal Understanding of MLLMs via Temporal-Aware Multi-Task Reinforcement Learning
Tao Wu, Li Yang, Gen Zhan, Yabin Zhang, Yiting Liao, Junlin Li, Deliang Fu, Li Zhang, Limin 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 instance number reward to TAL reward (ablation) | TempR1 · ActivityNet-v1.3 TAL | mF1: 69.8 -> 70.6 | Improved |
| use DP-based matching instead of sequential matching (ablation) | TempR1 · ActivityNet-v1.3 TAL | mF1: 45.4 -> 70.6 | Improved |
| GRPO instead of DAPO (ablation) | TempR1 · ActivityNet-v1.3 TAL | mF1: 47.6 -> 71.0 | Improved |
| GSPO instead of GRPO (ablation) | TempR1 · ActivityNet-v1.3 TAL | mF1: 71.0 -> 71.7 | No clear effect |
| add DTG and VHD tasks to TG+GVQA joint training (ablation) | TempR1 · QVHighlights VHD | mIoU: 7.1 -> 70.3 | Improved |
| add TAL task to joint training (ablation) | TempR1 · ActivityNet-v1.3 TAL | mF1: 68.2 -> 71.0 | Improved |
| add GVQA to TG-only training (ablation) | TempR1 · NExTGQA | Acc: 62.5 -> 67.5 | Improved |
| add DTG to TG+GVQA training (ablation) | TempR1 · ActivityNet DTG | mIoU: 17.7 -> 59.4 | Improved |
| add TAL to four-task training (ablation) | TempR1 · NExTGQA | Acc: 66.7 -> 70.1 | Improved |
| dataset-specific fine-tuning after multi-task RL | TempR1 · ActivityNet TG | mIoU: 32.2 -> 51.9 | Improved |
| RL fine-tuning instead of SFT on same data | TempR1 · general video benchmarks | accuracy | Improved |
| TempR1 multi-task vs VideoChat-R1 on Charades-STA | TempR1 · Charades-STA | mIoU: 60.8 -> 61.4 | Improved |
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 2 of 2 shared results
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
- TempSamp-R1: Effective Temporal Sampling with Reinforcement Fine-Tuning for Video LLMs (NeurIPS 2025 · similarity 0.90)
- Memory-T1: Reinforcement Learning for Temporal Reasoning in Multi-session Agents (ICLR 2026 · similarity 0.89)
- Video-R1: Reinforcing Video Reasoning in MLLMs (NeurIPS 2025 · similarity 0.88)
- Incentivizing Versatile Video Reasoning in MLLMs via Data-Efficient Reinforcement Learning (CVPR 2026 · similarity 0.88)
- TTRL: Test-Time Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- Find, Fix, Reason: Context Repair for Video Reasoning (ICML 2026 · similarity 0.87)
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.
