GarmentCodeData - 大规模3D定制服装合成数据集
收藏资源简介:
GarmentCodeData由苏黎世联邦理工学院与多特蒙德工业大学联合构建,是首个大规模的3D定制服装与缝纫图案的合成数据集。该数据集包含115,000个数据点,涵盖了多种设计,如上衣、衬衫、裙子、连体裤、裤子等,并适配多种基于CAESAR人体模型和标准参考人体形状的身体尺寸,同时应用了三种不同的纺织材料。为创建该数据集,研究者提出了一套算法,自动采集样本人体尺寸的裁缝测量数据,缝纫图案设计的采样策略,并提出了一个基于快速XPBD模拟器的开源3D服装披挂流程。该流程包括解决碰撞问题和确保披挂正确性的多种解决方案,以支持数据集的可扩展性。GarmentCodeData的应用领域广泛,从服装设计、虚拟试衣到3D服装重建等多个方面,为服装与时尚产业的创新发展提供了强有力的数据基础。
GarmentCodeData was jointly developed by ETH Zurich and Technische Universität Dortmund, and it is the first large-scale synthetic dataset of 3D custom garments and sewing patterns. This dataset contains 115,000 data points covering a variety of garment designs such as tops, shirts, skirts, jumpsuits, trousers and more. It supports a wide range of body sizes based on the CAESAR human body model and standard reference human body shapes, and incorporates three different textile materials. To construct this dataset, the researchers proposed a set of algorithms that automatically collect tailor's anthropometric measurement data for sampled human body sizes, a sampling strategy for sewing pattern design, as well as an open-source 3D garment draping pipeline based on the fast XPBD simulator. This pipeline incorporates multiple solutions for collision resolution and ensuring draping accuracy, which supports the scalability of the dataset. GarmentCodeData has broad application scenarios, including garment design, virtual try-on, 3D garment reconstruction and other fields, providing a solid data foundation for the innovative development of the apparel and fashion industries.




