Research map / Video / Video-language understanding
Video-HolmesV2: Can MLLMs Reason with Spatio-Temporal Audio-Visual Evidence in Long Videos?
Zhaoyang Wei, Zipeng Wang, Yushe Cao, Chenhui Qiang, Shuaibing Cheng, Xuesong Yang, Sen Nie, Bowen Jiang, Wenchao Ding, Yanchao Hao, Zheng Wei, Xuehui Yu and 1 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 |
|---|---|---|---|
| add text-guided compression to full tokens (ablation) | Ours (Qwen2.5-Omni-7B + token compression) · VideoMME | accuracy: 66.0 -> 67.0 | Improved |
| add audio-guided compression alone (ablation) | Ours (Qwen2.5-Omni-7B + token compression) · VideoMME | accuracy: 66.0 -> 63.3 | Hurt |
| add audio-guided compression to text-guided compression (ablation) | Ours (Qwen2.5-Omni-7B + token compression) · VideoMME | accuracy: 67.0 -> 67.4 | Improved |
| add dual-stream fusion and differentiated compression (ablation) | Ours (Qwen2.5-Omni-7B + token compression) · VideoMME | accuracy: 67.4 -> 67.8 | Improved |
| add dual-stream fusion and differentiated compression (ablation) | Ours (Qwen2.5-Omni-7B + token compression) · WorldScene | avg accuracy: 46.9 -> 47.2 | Improved |
| add text-guided compression to full tokens (ablation) | Ours (Qwen2.5-Omni-7B + token compression) · WorldScene | avg accuracy: 46.8 -> 46.5 | No clear effect |
| add audio-guided compression alone (ablation) | Ours (Qwen2.5-Omni-7B + token compression) · WorldScene | avg accuracy: 46.8 -> 45.1 | Hurt |
| increase token budget from 8192/2048 to 12288/4096 visual/audio (ablation) | Ours (Qwen2.5-Omni-7B + token compression) · VideoMME | accuracy: 67.8 -> 67.6 | Hurt |
| increase token budget from 8192/2048 to 12288/4096 visual/audio (ablation) | Ours (Qwen2.5-Omni-7B + token compression) · WorldScene | avg accuracy: 47.2 -> 47.3 | No clear effect |
| reduce budget to 4096 visual / 1024 audio tokens (ablation) | Ours (Qwen2.5-Omni-7B + token compression) · VideoMME | accuracy: 67.8 -> 66.5 | Hurt |
| use fixed-budget compression instead of full tokens (ablation) | Ours (Qwen2.5-Omni-7B + token compression) · ShortVid-Bench | avg score: 70.5 -> 70.8 | Improved |
| use fixed-budget compression instead of OmniZip 45% | Ours (Qwen2.5-Omni-7B + token compression) · AVUTBench | accuracy (avg): 63.0 -> 65.5 | 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.
- OmniZip 45% — OmniZip: Audio-Guided Dynamic Token Compression for Fast Omnimodal Large Language Models (CVPR 2026) · better on 3 of 3 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
- OmniVideoBench: Towards Audio-Visual Understanding Evaluation for Omni MLLMs (ICLR 2026 · similarity 0.91)
- MMOU: A Massive Multi-Task Omni Understanding and Reasoning Benchmark for Long and Complex Real-World Videos (NeurIPS 2026 · similarity 0.90)
- JointAVBench: A Benchmark for Joint Audio-Visual Reasoning Evaluation (ICLR 2026 · similarity 0.89)
- When Eyes and Ears Disagree: Can MLLMs Discern Audio-Visual Confusion? (AAAI 2026 · similarity 0.89)
- A Benchmark for Omni-Modal Reasoning in Long Videos (NeurIPS 2026 · similarity 0.89)
- StreamOV: Streaming Omni-Video Understanding via Evidence-Guided Memory and Response Triggering (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.
