IntPhys2 Benchmark

What is IntPhys2?

IntPhys2 tests intuitive physics—permanence, immutability, continuity and solidity—using synthetic videos. SOTA models perform around chance (~50 %), far below…

Released

2025-06-11

Evaluates

理解, Understanding, 视频生成, 物理智能, Embodied AI, 图像理解, Image Understanding, 视觉生成, Visual Generation, 空间理解, Spatial Understanding, 不支持, Unsupported

Openness

restricted

Importer

OpenCompass Hub

Reported scores

0

Source provenance

• Original evidence https://hub.opencompass.org.cn/dataset-detail/IntPhys2

IntPhys2 paper, code and dataset

• Paper https://arxiv.org/abs/2506.09849
• Code https://github.com/facebookresearch/IntPhys2
• Dataset https://huggingface.co/datasets/facebook/IntPhys2

Which sources cite IntPhys2?

• IntPhys2

Related 理解, Understanding, 视频生成, 物理智能, Embodied AI, 图像理解, Image Understanding, 视觉生成, Visual Generation, 空间理解, Spatial Understanding, 不支持, Unsupported benchmarks

• A-Bench
• A-OKVQA
• AbsPyramid
• ActionAtlas
• Ada-LEval
• AECBench
• AFQMC
• AgentBoard