Houses3K

该数据集名为Houses3K，包含3000个带有纹理的3D建筑模型，这些模型被分为12个批次，每个批次有50种不同的几何形状和5种不同的纹理变化。训练集由从6个选定批次中选取的256个对象组成，而测试集则由另一个批次中的50个对象组成。该数据集专注于表面重建，排除了具有复杂内部结构或多余底座的物体。规模上，该数据集共有3000个模型，其任务是对3D物体进行重建，并评估其泛化能力。

This dataset is named Houses3K. It encompasses 3,000 textured 3D architectural models, which are split into 12 batches. Each batch contains 50 unique geometric shapes and 5 distinct texture variants. The training set is composed of 256 objects selected from 6 designated batches, whereas the test set consists of 50 objects from a separate single batch. This dataset is tailored for surface reconstruction tasks, and objects with complex internal structures or redundant bases are excluded. In total, the dataset holds 3,000 models, with the core task being 3D object reconstruction and the evaluation of generalization performance.