Research map / Video / Video-language understanding
Seeing Is Believing: Grounding Long-Video Understanding in Spatio-Temporal Visual Evidence
Zhaoyang Wei, Guoliang Wang, Guohua Gao, Yanchao Hao, Mingda Li, Wenchao Ding, Xi Chen, Shizhu He, Xuehui 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 |
|---|---|---|---|
| add lightweight token selector (ablation) | DynamicSelect (Qwen2.5-VL-7B) · VideoMME Overall | accuracy: 65.4 -> 67.4 | Improved |
| add dynamic budget predictor (ablation) | DynamicSelect (Qwen2.5-VL-7B) · VideoMME Overall | accuracy: 67.4 -> 67.7 | Improved |
| add 5% merging of selected tokens (ablation) | DynamicSelect (Qwen2.5-VL-7B) · VideoMME Overall | accuracy: 67.7 -> 67.9 | Improved |
| raise merging ratio of selected tokens from 5% to 15% (ablation) | DynamicSelect (Qwen2.5-VL-7B) · VideoMME Overall | accuracy: 67.9 -> 67.2 | Hurt |
| add high-ratio compression (85%) of unselected high-score tokens (ablation) | DynamicSelect (Qwen2.5-VL-7B) · VideoMME Overall | accuracy: 67.9 -> 68.3 | Improved |
| auxiliary compression ratio 95% vs 85% (ablation) | DynamicSelect (Qwen2.5-VL-7B) · VideoMME Overall | accuracy: 68.3 -> 68.1 | Hurt |
| auxiliary compression ratio 75% vs 85% (ablation) | DynamicSelect (Qwen2.5-VL-7B) · VideoMME Overall | accuracy: 68.3 -> 68.3 | No clear effect |
| DynamicSelect vs FlexSelect (Qwen2.5-VL base) | DynamicSelect (Qwen2.5-VL-7B) · VideoMME Overall | accuracy: 67.4 -> 68.7 | Improved |
| DynamicSelect on Qwen2.5-VL-7B | DynamicSelect (Qwen2.5-VL-7B) · EV2-Bench | overall score: 44.8 -> 51.1 | Improved |
| DynamicSelect on Qwen2.5-VL-7B | DynamicSelect (Qwen2.5-VL-7B) · EV2-Bench Spatial | score: 59.0 -> 48.7 | Hurt |
| DynamicSelect on Qwen2.5-VL-7B | DynamicSelect (Qwen2.5-VL-7B) · EV2-Bench Temporal | score: 42.1 -> 62.1 | Improved |
| DynamicSelect on InternVL2.5-8B | DynamicSelect (Qwen2.5-VL-7B) · LVBench | accuracy: 43.4 -> 51.5 | 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.
- FlexSelect — FlexSelect: Flexible Token Selection for Efficient Long Video Understanding (NeurIPS 2025) · better on 7 of 7 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
- Before It Fades: Reinforcing Temporal Representations at Inference Time in VideoLLMs (NeurIPS 2026 · similarity 0.91)
- Temporal Chain of Thought: Long-Video Understanding by Thinking in Frames (NeurIPS 2025 · similarity 0.90)
- MMBench-Video: A Long-Form Multi-Shot Benchmark for Holistic Video Understanding (NeurIPS 2024 · similarity 0.90)
- Re-thinking Temporal Search for Long-Form Video Understanding (CVPR 2025 · similarity 0.90)
- Not Another Text Benchmark: Putting the “Visual" Back in Visual Question Answering for Large Video Models (NeurIPS 2026 · similarity 0.90)
- BLINK-Twice: You see, but do you observe? A Reasoning Benchmark on Visual Perception (NeurIPS 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.
