Research map / Video / Video-language understanding
Benchmarking MLLMs on Mistake Recognition and Explanation in Single-Step Components of Cooking
Shun Takashige, Atsushi Hashimoto, Shin’ichi Satoh
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 type-specific hint prompts (ablation) | InternVL3.5-38B · Cook-MRE | Correction accuracy (%): 50.2 -> 57.4 | Improved |
| add type-specific hint prompts (ablation) | VTimeLLM-7B · Cook-MRE | Mistake presence F1 (%): 22.4 -> 58.8 | Mixed |
| add type-specific hint prompts (ablation) | ST-LLM-7B · Cook-MRE | Mistake presence F1 (%): 80.3 -> 86.8 | Mixed |
| add type-specific hint prompts (ablation) | ST-LLM-7B · Cook-MRE | Full Pass accuracy (%): 10.7 -> 2.0 | Hurt |
| add type-specific hint prompts (ablation) | VTimeLLM-7B · Cook-MRE | Full Pass accuracy (%): 22.7 -> 10.6 | Hurt |
| add type-specific hint prompts (ablation) | Tarsier-34B · Cook-MRE | Full Pass accuracy (%): 39.6 -> 36.3 | Hurt |
| add type-specific hint prompts (ablation) | InternVL3.5-38B · Cook-MRE | Full Pass accuracy (%): 55.5 -> 62.1 | Improved |
| add type-specific hint prompts (ablation) | Qwen2.5-VL-32B · Cook-MRE | Mistake presence F1 (%): 77.8 -> 74.0 | Hurt |
| scale InternVL3 from 8B to 38B (ablation) | InternVL3 · Cook-MRE verb easy | Full Pass accuracy (%): +35.5 | Improved |
| scale InternVL3 from 8B to 38B (ablation) | InternVL3 · Cook-MRE verb hard | Full Pass accuracy (%): +12.4 | Improved |
| scale InternVL3 from 8B to 38B (ablation) | InternVL3 · Cook-MRE ingredient hard | Full Pass accuracy (%): +17.4 | Improved |
| scale Qwen2.5-VL from 7B to 32B (ablation) | Qwen2.5-VL · Cook-MRE hard problems | Full Pass accuracy (%) | No clear effect |
Closest papers by meaning
- CaptainCook4D: A Dataset for Understanding Errors in Procedural Activities (NeurIPS 2024 · similarity 0.88)
- PREGO: Online Mistake Detection in PRocedural EGOcentric Videos (CVPR 2024 · similarity 0.83)
- Error Detection in Egocentric Procedural Task Videos (CVPR 2024 · similarity 0.81)
- Modeling Multiple Normal Action Representations for Error Detection in Procedural Tasks (CVPR 2025 · similarity 0.81)
- Can Multi-Modal LLMs Provide Live Step-by-Step Task Guidance? (NeurIPS 2025 · similarity 0.81)
- SLVMBench: Skill Learning from Video Memory (NeurIPS 2026 · similarity 0.81)
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.
