Physion

该数据集由ThreeDWorld模拟器创建，规模宏大，包含八个不同的场景（多米诺、容器、碰撞、掉落、滚动、连接、支撑和悬挂），展现了丰富多样的物理场景和复杂的物体交互。此外，该数据集还为可变形物体的场景提供了具有挑战性的基准，用于评估动力学模型。其规模属于大型，主要任务是对动力学模型进行评估。This dataset is generated using the ThreeDWorld simulator. It has a large scale and includes eight distinct scenarios: Domino, Container, Collision, Falling, Rolling, Connecting, Supporting, and Hanging, showcasing a wide range of physical scenes and complex object interactions. Additionally, this dataset offers challenging benchmarks for scenarios involving deformable objects, which are designed to evaluate dynamic models. As a large-scale dataset, its primary purpose is to evaluate dynamic models.

搜集汇总

背景与挑战

背景概述

Physion是一个用于评估人类和机器视觉物理预测能力的基准数据集，源自NeurIPS 2021论文。它包含八种物理场景（如碰撞、支撑、多米诺等）的训练和测试数据，提供视频、分割图和HDF5文件格式，旨在支持模型训练和人类行为实验的比较分析。