Boxhead

Boxhead是由马克斯普朗克智能系统研究所创建的一个新数据集，旨在学习具有层次结构表示的数据。该数据集包含一个简单的3D场景中的立方体，具有10个离散的变化因素，如墙色、地板色、立方体色等。数据集通过Open3D和PyVista创建，包含三个变体，每个变体展示了不同程度的微观和宏观因素之间的依赖关系。Boxhead数据集的应用领域包括评估和改进基于变分自编码器的解耦模型，特别是在处理真实世界数据的层次结构特性方面。

Boxhead is a novel dataset developed by the Max Planck Institute for Intelligent Systems, designed to enable models to learn data with hierarchical representations. This dataset includes cubes within a simple 3D scene, featuring 10 discrete variation factors such as wall color, floor color, cube color, and others. Constructed using Open3D and PyVista, the dataset encompasses three variants, each demonstrating distinct degrees of dependencies between microscopic and macroscopic factors. The application scenarios of the Boxhead dataset include evaluating and improving variational autoencoder-based disentangled models, particularly when addressing the hierarchical structural properties of real-world data.