Visual Persona-500K
收藏资源简介:
Visual Persona-500K是由韩国科学技术院(KAIST AI)和Adobe Research创建的大型数据集,包含580k张配对的人体图像,涵盖100k个独特身份。该数据集通过视觉语言模型评估人体外观一致性,生成详细文本描述,以区分个体身份和图像内的变化。数据集广泛应用于全身人体定制领域,解决了文本对齐和身份保持两大关键问题,可应用于虚拟试穿、人体风格化和角色定制等多种应用场景。
Visual Persona-500K is a large-scale dataset created by the Korea Advanced Institute of Science and Technology (KAIST AI) and Adobe Research. It contains 580k paired human images covering 100k unique identities. This dataset uses vision-language models to assess the consistency of human appearance, and generates detailed textual descriptions to distinguish individual identities and intra-image variations. Widely applied in the field of full-body human customization, the dataset addresses two key challenges: text alignment and identity preservation. It can be deployed in various application scenarios such as virtual try-on, human stylization and character customization.

- 1Visual Persona: Foundation Model for Full-Body Human Customization韩国科学技术院(KAIST AI), Adobe Research · 2025年



