USC-GVL/PhysBench
收藏Hugging Face2025-03-05 更新2025-04-12 收录
下载链接：
https://hf-mirror.com/datasets/USC-GVL/PhysBench
下载链接
链接失效反馈官方服务：
资源简介：
PhysBench是一个用于评估视觉语言模型在理解物理世界方面的能力的综合基准。它包含四个主要领域：物理对象属性、物理对象关系、物理场景理解和基于物理的动力学，进一步细分为19个子类和8个不同的能力维度。数据集由10002个测试条目和200个验证条目组成，测试条目更具挑战性和多样性。
PhysBench is a comprehensive benchmark designed to evaluate Vision-Language Models (VLMs) capability in understanding the physical world across a diverse set of tasks. It encompasses four major domains: physical object properties, physical object relationships, physical scene understanding, and physics-based dynamics, further divided into 19 subclasses and 8 distinct capability dimensions. The dataset consists of 10,002 test entries and 200 validation entries, with the test set being more challenging and diverse.
提供机构：
USC-GVL
