Research map / Video / Video-language understanding
MIMIC-Bench: Exploring the User-Like Thinking and Mimicking Capabilities of Multimodal Large Language Models
Jiajie Teng, Huiyu Duan, Sijing Wu, Jiarui Wang, Xilei Zhu, Jianing Jin, Wei Shen, Xiongkuo Min, Guangtao Zhai
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 temporal encoder branch (ablation) | MIMIC-Chat · MIMIC-Bench thinking | Overall accuracy (%): 65.8 -> 74.1 | Improved |
| add temporal encoder branch (ablation) | MIMIC-Chat · MIMIC-Bench DeS | accuracy (%): 53.3 -> 87.1 | Improved |
| add LoRA adaptation (ablation) | MIMIC-Chat · MIMIC-Bench thinking | Overall accuracy (%): 66.9 -> 74.1 | Improved |
| add LoRA adaptation (ablation) | MIMIC-Chat · MIMIC-Bench TaM | accuracy (%): 70.4 -> 86.7 | Improved |
| add dedicated temporal projector (ablation) | MIMIC-Chat · MIMIC-Bench thinking | Overall accuracy (%): 67.5 -> 74.1 | Improved |
| add dedicated spatial projector (ablation) | MIMIC-Chat · MIMIC-Bench thinking | Overall accuracy (%): 68.7 -> 74.1 | Improved |
| add temporal encoder branch (ablation) | MIMIC-Chat · MIMIC-Bench CoP | accuracy (%): 32.6 -> 43.6 | Improved |
| fine-tune on MIMIC-Data | Qwen2.5-VL-7B · MIMIC-Bench thinking | Overall accuracy (%): 59.9 -> 66.4 | Improved |
| fine-tune on MIMIC-Data | InternVideo2.5-8B · MIMIC-Bench thinking | Overall accuracy (%): 66.3 -> 68.1 | Improved |
| fine-tune on MIMIC-Data | InternVL2.5-8B · MIMIC-Bench thinking | Overall accuracy (%): 64.5 -> 65.8 | Improved |
| use MIMIC-Chat architecture instead of fine-tuned existing MLLM | MIMIC-Data fine-tuning · MIMIC-Bench thinking | Overall accuracy (%): 68.1 -> 74.1 | Improved |
| scale model from Qwen2.5-VL 7B to 72B | Qwen2.5-VL · MIMIC-Bench comment simulation | Judged as Human (%): 10.25 -> 26.93 | Improved |
Closest papers by meaning
- MMOU: A Massive Multi-Task Omni Understanding and Reasoning Benchmark for Long and Complex Real-World Videos (NeurIPS 2026 · similarity 0.87)
- Video-Only ToM: Enhancing Theory of Mind in Multimodal Large Language Models (CVPR 2026 · similarity 0.87)
- MMBench-Video: A Long-Form Multi-Shot Benchmark for Holistic Video Understanding (NeurIPS 2024 · similarity 0.87)
- EgoExoBench: A Benchmark for First- and Third-person View Video Understanding in MLLMs (NeurIPS 2025 · similarity 0.87)
- MMWorld: Towards Multi-discipline Multi-faceted World Model Evaluation in Videos (ICLR 2025 · similarity 0.87)
- MMR-V: What's Left Unsaid? A Benchmark for Multimodal Deep Reasoning in Videos (ICLR 2026 · similarity 0.87)
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.
