Research map / Video / Video-language understanding
From Trial to Triumph: Advancing Long Video Understanding via Visual Context Sample Scaling and Self-reward Alignment
Yucheng Suo, Fan Ma, Linchao Zhu, Tianyi Wang, Fengyun Rao, Yi Yang
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 marginal confidence score S^mc to frequency score (ablation) | Llava-Video-7B · MLVU | accuracy: 69.37 -> 71.17 | Improved |
| add voting score S^v on top of S^f and S^mc (ablation) | Llava-Video-7B · Video-MME | accuracy: 64.48 -> 66.52 | Improved |
| add marginal confidence score S^mc to frequency score (ablation) | VideoLLaMA2-7B · Video-MME | accuracy: 49.11 -> 49.81 | Improved |
| add voting score S^v on top of S^f and S^mc (ablation) | VideoLLaMA2-7B · MLVU | accuracy: 53.98 -> 55.31 | Improved |
| use Deepseek-R1 instead of Llama-3.3-70B as external LLM (ablation) | Llava-Video-7B · Video-MME | accuracy: 66.30 -> 66.52 | Improved |
| use marginal confidence instead of max logit (intra-sample) (ablation) | Llava-Video-7B · Video-MME | accuracy: 66.26 -> 66.52 | Improved |
| use max instead of mean across samples (inter-sample) (ablation) | Llava-Video-7B · Video-MME | accuracy: 66.22 -> 66.52 | Improved |
| reduce frames per sample from 32 to 8 (ablation) | Llava-Video-7B · Video-MME | accuracy: 66.52 -> 62.48 | Hurt |
| method gain over baseline at 8 frames (ablation) | Llava-Video-7B · Video-MME (8 frames) | accuracy: 56.26 -> 62.48 | Improved |
| method gain over baseline at 16 frames (ablation) | Llava-Video-7B · Video-MME (16 frames) | accuracy: 60.18 -> 63.45 | Improved |
| bin-wise instead of fully random frame sampling (ablation) | Llava-Video-7B · Video-MME | majority accuracy: +2% | Improved |
| use 32 instead of 8 frames per sample for majority voting (ablation) | Llava-Video-7B · Video-MME | majority accuracy: over 5% | Improved |
Closest papers by meaning
- Temporal Chain of Thought: Long-Video Understanding by Thinking in Frames (NeurIPS 2025 · similarity 0.94)
- LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding (NeurIPS 2024 · similarity 0.93)
- CG-Bench: Clue-grounded Question Answering Benchmark for Long Video Understanding (ICLR 2025 · similarity 0.92)
- Not Another Text Benchmark: Putting the “Visual" Back in Visual Question Answering for Large Video Models (NeurIPS 2026 · similarity 0.92)
- Streaming Long Video Understanding with Large Language Models (NeurIPS 2024 · similarity 0.92)
- ReWind: Understanding Long Videos with Instructed Learnable Memory (CVPR 2025 · similarity 0.91)
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.
