FIT (Fit-Inclusive Try-on)
收藏资源简介:
FIT是由华盛顿大学和谷歌研究院联合创建的大规模虚拟试衣数据集,专注于服装合身性评估。该数据集包含113万组训练样本,涵盖168种体型、528种姿势及15.8万种服装设计,通过GarmentCode程序化生成3D服装并模拟真实物理垂坠效果。数据采用合成到真实(sim2real)的纹理重建技术,确保几何精度与照片级真实性。其核心价值在于解决现有试衣系统忽略尺寸匹配的问题,为电商、时尚设计等领域提供精准的虚拟合身性预测基准。
FIT is a large-scale virtual try-on dataset jointly created by the University of Washington and Google Research, focusing on clothing fit assessment. It contains 1.13 million training samples, encompassing 168 body types, 528 poses, and 158,000 clothing designs. 3D garments are programmatically generated via GarmentCode, with realistic physical draping effects simulated. The dataset adopts synthetic-to-real (sim2real) texture reconstruction techniques to ensure geometric accuracy and photorealistic authenticity. Its core value lies in addressing the issue that existing try-on systems overlook size matching, providing an accurate virtual fit prediction benchmark for industries such as e-commerce and fashion design.

- 1FIT: A Large-Scale Dataset for Fit-Aware Virtual Try-On华盛顿大学; 谷歌研究院 · 2026年



