Research map / Video / Video-language understanding
VideoARM: Agentic Reasoning over Hierarchical Memory for Long-Form Video Understanding
Yufei Yin, Qianke Meng, Minghao Chen, Jiajun Ding, Zhenwei Shao, Zhou Yu
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 |
|---|---|---|---|
| replace DVD with VideoARM (same o3+GPT-4.1) | long-video agent · Video-MME long | accuracy (%): 67.3 -> 75.3 | Improved |
| replace DVD with VideoARM | long-video agent · LVBench | accuracy (%): 76.0 -> 79.7 | Improved |
| use VideoARM instead of DVD for token usage | long-video agent · Video-MME (10 videos, 30 queries) | tokens: 64.21M -> 1.89M | Improved |
| add Scene Snapper (ablation) | VideoARM · Video-MME long (200-QA subset) | accuracy (%): 69.0 -> 76.5 | Improved |
| add Audio Transcriber (ablation) | VideoARM · Video-MME long (200-QA subset) | accuracy (%): 70.0 -> 76.5 | Improved |
| add Clip Analyzer (ablation) | VideoARM · Video-MME long (200-QA subset) | accuracy (%): 75.5 -> 76.5 | Improved |
| add long-term perception pool (ablation) | VideoARM · Video-MME long (200-QA subset) | accuracy (%): 67.0 -> 76.5 | Improved |
| add short-term perception pool (ablation) | VideoARM · Video-MME long (200-QA subset) | accuracy (%): 72.5 -> 76.5 | Improved |
| add Working Memory (ablation) | VideoARM · Video-MME long (200-QA subset) | accuracy (%): 75.5 -> 76.5 | Improved |
| add HM3 instead of controller context only (ablation) | VideoARM · Video-MME long (200-QA subset) | accuracy (%): 74.5 -> 76.5 | Improved |
| use o3 instead of GPT-4o as controller (ablation) | VideoARM · Video-MME long (200-QA subset) | accuracy (%): 40.5 -> 76.5 | Improved |
| use GPT-4o instead of GPT-4.1 in understanding tools (ablation) | VideoARM (o3) · Video-MME long (200-QA subset) | accuracy (%): 76.5 -> 80.0 | 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.
- VideoLucy — VideoLucy: Deep Memory Backtracking for Long Video Understanding (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
- LongVT: Incentivizing "Thinking with Long Videos" via Native Tool Calling (CVPR 2026 · similarity 0.90)
- VideoSeek: Long-Horizon Video Agent with Tool-Guided Seeking (CVPR 2026 · similarity 0.90)
- Seeing, Listening, Remembering, and Reasoning: A Multimodal Agent with Long-Term Memory (ICLR 2026 · similarity 0.89)
- ReAgent-V: A Reward-Driven Multi-Agent Framework for Video Understanding (NeurIPS 2025 · similarity 0.89)
- ViLoMem: Agentic Learner with Grow-and-Refine Multimodal Semantic Memory (CVPR 2026 · similarity 0.88)
- Video-R1: Reinforcing Video Reasoning in MLLMs (NeurIPS 2025 · 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.
