Research map / Video / Video-language understanding
Evidence-Backed Video Question Answering
Shijie Wang, Honglu Zhou, Ziyang Wang, Ran Xu, Caiming Xiong, Silvio Savarese, Chen Sun, Juan Carlos Niebles
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 |
|---|---|---|---|
| fine-tune UniPixel-7B on ST-Evidence-Instruct | UniPixel-7B · ST-Evidence-Gen | t-mean: 1.9 -> 29.1 | Improved |
| fine-tune UniPixel-7B on ST-Evidence-Instruct | UniPixel-7B · ST-Evidence-Gen | J&F: 40.3 -> 54.1 | Improved |
| fine-tune UniPixel-3B on ST-Evidence-Instruct | UniPixel-3B · ST-Evidence-Gen | t-mean: 6.5 -> 26.9 | Improved |
| fine-tune UniPixel-3B on ST-Evidence-Instruct | UniPixel-3B · ST-Evidence-Gen | QA accuracy: 72.2 -> 72.4 | No clear effect |
| fine-tune UniPixel-7B on ST-Evidence-Instruct | UniPixel-7B · MeViS_u | J&F: 61.7 -> 62.5 | Improved |
| fine-tune UniPixel-3B on ST-Evidence-Instruct | UniPixel-3B · MVBench | accuracy: 62.5 -> 62.3 | No clear effect |
| corrupt annotated evidence regions (ablation) | Gemini-2.5-Flash · ST-Evidence-Gen (200-sample subset) | QA accuracy: 81.32 -> 61.12 | Hurt |
| corrupt non-evidence regions (ablation) | Gemini-2.5-Flash · ST-Evidence-Gen (200-sample subset) | QA accuracy: 81.32 -> 76.43 | Hurt |
| corrupt annotated evidence regions (ablation) | Qwen3-VL-8B · ST-Evidence-Gen (200-sample subset) | QA accuracy: 79.31 -> 58.32 | Hurt |
| replace Qwen2.5-VL-72B with the smaller Qwen3-VL-4B | Qwen-VL family · ST-Evidence-Gen | t-mean: 25.3 -> 38.6 | Improved |
| scale model size within a family | Qwen2.5-VL / Qwen3-VL · ST-Evidence-Gen | QA accuracy vs evidence | Mixed |
| use manual referring expressions instead of Gemini-2.5-Pro expressions with proxy segmenter (ablation) | UniPixel-3B · ST-Evidence-Gen (100 samples) | J&F: 25.6 -> 44.7 | Improved |
Closest papers by meaning
- VITED: Video Temporal Evidence Distillation (CVPR 2025 · similarity 0.92)
- Commonsense Video Question Answering through Video-Grounded Entailment Tree Reasoning (CVPR 2025 · similarity 0.92)
- CG-Bench: Clue-grounded Question Answering Benchmark for Long Video Understanding (ICLR 2025 · similarity 0.92)
- LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding (NeurIPS 2024 · similarity 0.92)
- Can I Trust Your Answer? Visually Grounded Video Question Answering (CVPR 2024 · similarity 0.91)
- Temporal Chain of Thought: Long-Video Understanding by Thinking in Frames (NeurIPS 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.
