Research map / Video / Video-language understanding
MMDuet2: Enhancing Proactive Interaction of Video MLLMs with Multi-Turn Reinforcement Learning
Yueqian Wang, Songxiang Liu, Disong Wang, Nuo Xu, Wan Guanglu, Huishuai Zhang, Dongyan Zhao
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 multi-turn GRPO RL with PAUC-based reward after SFT | MMDuet2 SFT · ProactiveVideoQA [WEB] | PAUC: 37.6 -> 53.3 | Improved |
| add RL after SFT | MMDuet2 SFT · ProactiveVideoQA [EGO] | PAUC: 26.4 -> 33.6 | Improved |
| add RL after SFT | MMDuet2 SFT · StreamingBench Proactive Output | accuracy: 19.59 -> 34.69 | Improved |
| add replication reward r_rep (ablation) | MMDuet2 · ProactiveVideoQA [WEB] | duplicate proportion: 17.3 -> 4.2 | Improved |
| add in-span reward r_in_span (ablation) | MMDuet2 · ProactiveVideoQA [EGO] | feasibility | Improved |
| add prefix reward r_pfx (ablation) | MMDuet2 · ProactiveVideoQA [EGO] | PAUC: 27.5 -> 33.6 | Improved |
| add r_pfx (ablation) | MMDuet2 · ProactiveVideoQA [WEB] | PAUC: 53.0 -> 53.3 | No clear effect |
| reduce inference frame interval from 2 s to 1 s (ablation) | MMDuet2 RL · ProactiveVideoQA [WEB] | PAUC: 44.2 -> 53.3 | Improved |
| SFT frame interval 1 s instead of 2 s (ablation) | MMDuet2 SFT · ProactiveVideoQA | behavior | Hurt |
| RL frame interval 1 s vs 2 s (ablation) | MMDuet2 RL · ProactiveVideoQA [EGO] | PAUC: 33.6 -> 34.7 | No clear effect |
| proactive SFT+RL training | Qwen2.5-VL 3B · Video-MME (w/o sub) | accuracy: 57.3 -> 58.1 | No clear effect |
| proactive SFT+RL training | Qwen2.5-VL 3B · LongVideoBench | accuracy: 53.1 -> 52.7 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Dispider — Dispider: Enabling Video LLMs with Active Real-Time Interaction via Disentangled Perception, Decision, and Reaction (CVPR 2025) · better on 1 of 1 shared result
- VideoLLM-Online — VideoLLM-online: Online Video Large Language Model for Streaming Video (CVPR 2024) · 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
- IPIBench: Evaluating Interactive Proactive Intelligence of MLLMs under Continuous Streams (NeurIPS 2026 · similarity 0.89)
- Incentivizing Versatile Video Reasoning in MLLMs via Data-Efficient Reinforcement Learning (CVPR 2026 · similarity 0.87)
- MMDU: A Multi-Turn Multi-Image Dialog Understanding Benchmark and Instruction-Tuning Dataset for LVLMs (NeurIPS 2024 · similarity 0.87)
- Video-R1: Reinforcing Video Reasoning in MLLMs (NeurIPS 2025 · similarity 0.86)
- Streaming Video Understanding and Multi-round Interaction with Memory-enhanced Knowledge (ICLR 2025 · similarity 0.86)
- SAMA: Towards Multi-Turn Referential Grounded Video Chat with Large Language Models (NeurIPS 2025 · similarity 0.86)
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.
