Research map / Video / Video-language understanding
Q-Frame: Query-aware Frame Selection and Multi-Resolution Adaptation for Video-LLMs
Shaojie Zhang, Jiahui Yang, Jianqin Yin, Zhenbo Luo, Jian Luan
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 Q-Frame query-aware selection and multi-resolution to Qwen2-VL | Qwen2-VL · MLVU | accuracy: 56.9 -> 65.4 | Improved |
| add Q-Frame to GPT-4o frame selection | GPT-4o · LongVideoBench val | accuracy: 53.3 -> 58.6 | Improved |
| add Q-Frame frame selection to VILA-V1.5 instead of Frame-Voyager | VILA-V1.5 · MLVU | accuracy: 49.8 -> 54.4 | Improved |
| add Q-Frame to VILA-V1.5 on very short videos (8s,15s] | VILA-V1.5 · LongVideoBench (8s,15s] | accuracy: 56.1 -> 52.9 | Hurt |
| add Q-Frame to Qwen2-VL on long videos (15m,60m] | Qwen2-VL · LongVideoBench (15m,60m] | accuracy: 45.9 -> 56.4 | Improved |
| add Gumbel-Max QFS sampling to CLIP top-K selection (ablation) | Q-Frame · LongVideoBench val | accuracy: 56.0 -> 57.6 | Improved |
| add query-aware QFS selection to MRA (ablation) | Q-Frame · LongVideoBench val | accuracy: 52.6 -> 58.4 | Improved |
| add MRA multi-resolution to QFS (ablation) | Q-Frame · LongVideoBench val | accuracy: 57.6 -> 58.4 | Improved |
| add medium and low resolution frames to high-only (ablation) | Q-Frame MRA · LongVideoBench val | accuracy: 57.6 -> 58.4 | Improved |
| use only low-resolution frames instead of mixed resolutions (ablation) | Q-Frame MRA · LongVideoBench val | accuracy: 58.4 -> 49.0 | Hurt |
| increase low-resolution frames from 32 to 48 (ablation) | Q-Frame MRA · LongVideoBench val | accuracy: 58.4 -> 57.4 | Hurt |
| use Long-CLIP instead of standard CLIP for retrieval (ablation) | Q-Frame CQR · LongVideoBench val | accuracy: 57.8 -> 58.4 | Improved |
Closest papers by meaning
- M-LLM Based Video Frame Selection for Efficient Video Understanding (CVPR 2025 · similarity 0.91)
- Adaptive Keyframe Sampling for Long Video Understanding (CVPR 2025 · similarity 0.91)
- Frame-Voyager: Learning to Query Frames for Video Large Language Models (ICLR 2025 · similarity 0.91)
- BOLT: Boost Large Vision-Language Model Without Training for Long-form Video Understanding (CVPR 2025 · similarity 0.90)
- Logic-in-Frames: Dynamic Keyframe Search via Visual Semantic-Logical Verification for Long Video Understanding (NeurIPS 2025 · similarity 0.89)
- SlowFocus: Enhancing Fine-grained Temporal Understanding in Video LLM (NeurIPS 2024 · 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.
