VKnowU Benchmark

What is VKnowU?

尽管多模态大型语言模型（MLLMs）在识别物体方面已相当熟练，但它们往往缺乏对世界潜在物理和社会原则的直觉的、类似人类的理解。这种高级的、基于视觉的语义，我们称之为视觉知识，其构成了感知与推理之间的桥梁。为了系统地评估这种能力，我们提出了 VKnowU，一个包含 1,249 个视频、1,680…

Released

2025-11-25

Modality

multimodal

Evaluates

多模态, Multimodal, 推理, Reasoning, 视频理解, Video-Understanding, Visual Knowledge, Physics, Psychology, 物理智能, Embodied AI, 逻辑推理, Video Understanding, 不支持, Unsupported

Openness

restricted

Importer

OpenCompass Hub

Reported scores

0

Source provenance

• Original evidence https://hub.opencompass.org.cn/dataset-detail/VKnowU

VKnowU paper, code and dataset

• Paper https://img.shields.io/badge/Paper-arXiv:2511.20272-b31b1b.svg
• Code https://github.com/OpenGVLab/VKnowU
• Dataset https://huggingface.co/datasets/Eurayka/VKnowU
• Dataset https://huggingface.co/datasets/OpenGVLab/VKnowU

Which sources cite VKnowU?

• VKnowU

Related 多模态, Multimodal, 推理, Reasoning, 视频理解, Video-Understanding, Visual Knowledge, Physics, Psychology, 物理智能, Embodied AI, 逻辑推理, Video Understanding, 不支持, Unsupported benchmarks

• 3DCodeBench
• A-Bench
• A-OKVQA
• AbsPyramid
• ActionAtlas
• ActiveFly-Bench
• Ada-LEval
• AdvancedMathBench