Text-to-3D House Model (Text--to--3D House Model)

该数据集包含 2,000 个房屋、13,478 个房间和 873 个（一些房间具有相同的纹理，因此这个数字小于房间的总数。）具有相应自然语言描述的纹理图像。这些描述首先由一些预定义的模板生成，然后由人工提炼。描述的平均长度为 173.73，有 193 个唯一词。在我们的实验中，我们使用 1,600 对进行训练，而在生成建筑布局时使用 400 对进行测试。对于纹理合成，我们使用 503 个数据进行训练和 370 个数据进行测试。

This dataset contains 2,000 houses, 13,478 rooms, and 873 textured images with corresponding natural language descriptions (some rooms share identical textures, thus this count is less than the total number of rooms). These descriptions were initially generated via predefined templates, followed by manual refinement. The average length of the descriptions is 173.73, with a total of 193 unique words. In our experiments, 1,600 pairs were utilized for training, while 400 pairs were used for testing during architectural layout generation. For the texture synthesis task, 503 samples were allocated for training and 370 for testing.

搜集汇总

背景与挑战

背景概述

该数据集包含2000个房屋、13478个房间和873个纹理图像，每个都配有自然语言描述，这些描述由模板生成后经人工提炼，平均长度为173.73词。数据用于从语言描述生成3D房屋设计的实验，包括建筑布局和纹理合成的训练与测试部分。