Research map / Video / Video-language understanding
A.I.R.: Enabling Adaptive, Iterative, and Reasoning-based Frame Selection For Video Question Answering
Yuanhao Zou, Shengji Jin, Andong Deng, Youpeng Zhao, Jun Wang, Chen Chen
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 A.I.R. frame selection over uniform sampling | QwenVL-2.5-7B · Video-MME w/o subtitles | accuracy (%): 60.8 -> 65.0 | Improved |
| add A.I.R. frame selection over uniform sampling | QwenVL-2.5-7B · NextQA | accuracy (%): 74.3 -> 81.3 | Improved |
| use A.I.R. instead of MDP3 frame selection | InternVL-3-8B · MLVU dev | accuracy (%): 74.0 -> 74.5 | Improved |
| add A.I.R. frame selection over uniform sampling | InternVL-3-8B · NextQA | accuracy (%): 82.3 -> 82.6 | Improved |
| add A.I.R. frame selection over uniform sampling | VILA-1.5-8B · Video-MME w/o subtitles | accuracy (%): 48.9 -> 53.7 | Improved |
| add A.I.R. over MDP3 | QwenVL-2.5-7B · Egoschema Full | accuracy (%): 56.8 -> 58.8 | Improved |
| raise max frame budget to 256 with A.I.R. versus uniform 256 (ablation) | QwenVL-2.5-7B · Video-MME w/o subtitles | accuracy (%): 63.4 -> 66.4 | Improved |
| use A.I.R. in place of BOLT at 32 frames | LLaVA-OneVision-7B · Video-MME w/o subtitles | accuracy (%): 59.9 -> 61.4 | Improved |
| use A.I.R. versus BOLT on Egoschema subset | LLaVA-OneVision-7B · Egoschema Subset (32 frames) | accuracy (%): 64.0 -> 63.2 | Mixed |
| scale base VLM from 7B to larger QwenVL-2.5 models (ablation) | A.I.R. · QwenVL-2.5 7B/32B/72B | accuracy gain: +4.2% | Mixed |
| reduce worst-case VLM workload w_worst from 72 to 16 (ablation) | A.I.R. · NextQA | accuracy (%): 82.6 -> 81.7 | Hurt |
| remove reasoning-based VLM analysis (ablation) | A.I.R. · component ablation | accuracy | Improved |
Closest papers by meaning
- Flexible Frame Selection for Efficient Video Reasoning (CVPR 2025 · similarity 0.91)
- VideoEspresso: A Large-Scale Chain-of-Thought Dataset for Fine-Grained Video Reasoning via Core Frame Selection (CVPR 2025 · similarity 0.91)
- Commonsense Video Question Answering through Video-Grounded Entailment Tree Reasoning (CVPR 2025 · similarity 0.90)
- Frame-Voyager: Learning to Query Frames for Video Large Language Models (ICLR 2025 · similarity 0.90)
- Temporal Chain of Thought: Long-Video Understanding by Thinking in Frames (NeurIPS 2025 · similarity 0.90)
- M-LLM Based Video Frame Selection for Efficient Video Understanding (CVPR 2025 · similarity 0.90)
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.
