Research map / Video / Video-language understanding
ReQuest: Rethinking-based Question-Aware Frame Selection for Long-Form Video QA
Minkuk Kim, Suyong Yun, Young Kim, Jinyoung Moon, Jinwoo Choi, Seong Tae Kim
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 ReQuest frame selection | LLaVA-Video 7B · Video-MME w/o subtitles overall | accuracy: 62.6 -> 65.6 | Improved |
| add ReQuest frame selection | LLaVA-Video 7B · Video-MME w/o subtitles long | accuracy: 52.2 -> 55.8 | Improved |
| add ReQuest frame selection | LLaVA-Video 7B · MLVU m-avg | accuracy: 64.7 -> 71.7 | Improved |
| add ReQuest frame selection | Qwen3-VL-32B · Video-MME w/o subtitles overall | accuracy: 73.9 -> 75.7 | Improved |
| add ReQuest frame selection | Qwen3-VL-8B · LongVideoBench val | accuracy: 62.7 -> 66.3 | Improved |
| use adaptive NMS instead of greedy NMS (ablation) | ReQuest on LLaVA-Video · Video-MME w/o subtitles long | accuracy: 53.3 -> 55.8 | Improved |
| use adaptive NMS instead of top-k (ablation) | ReQuest on LLaVA-Video · Video-MME w/o subtitles overall | accuracy: 64.0 -> 65.6 | Improved |
| add uncertainty routing (1365 selection / 1335 uniform) instead of routing all to selection (ablation) | ReQuest on LLaVA-Video · Video-MME w/o subtitles overall | accuracy: 65.3 -> 65.6 | Improved |
| route everything to uniform sampling (no selection) (ablation) | ReQuest on LLaVA-Video · Video-MME w/o subtitles overall | accuracy: 65.6 -> 62.6 | Hurt |
| train selector with MLLM response supervision instead of model-based similarity (ablation) | ReQuest selector BLIP · Video-MME w/o subtitles overall | accuracy: 63.6 -> 65.6 | Improved |
| use frame-level instead of segment-level inference (ablation) | ReQuest on LLaVA-Video · Video-MME w/o subtitles overall | accuracy: 62.9 -> 65.6 | Improved |
| replace uniform re-inference with selection re-inference (ablation) | Router-Ours · Video-MME w/o subtitles overall | accuracy: 62.8 -> 65.6 | 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.
- LongVU — LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understanding (ICML 2025) · 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
- LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding (NeurIPS 2024 · similarity 0.91)
- Flexible Frame Selection for Efficient Video Reasoning (CVPR 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)
- Adaptive Keyframe Sampling for Long Video Understanding (CVPR 2025 · similarity 0.90)
- BIMBA: Selective-Scan Compression for Long-Range Video Question Answering (CVPR 2025 · similarity 0.89)
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.
