CelebA-Dialog (HQ)

CelebA-Dialog是一个大规模的视觉语言人脸数据集，具有以下特征:
面部图像带有丰富的细粒度标签，该标签根据一个属性的语义含义将其分为多个等级。
与每个图像一起，都有描述属性的标题和用户请求示例。
总部:
通过遵循CelebA-HQ从CelebA数据集中选择30,000面部图像
1024x1024的高分辨率

CelebA-Dialog is a large-scale vision-language facial dataset with the following characteristics:
Facial images are accompanied by rich fine-grained labels, which are categorized into multiple levels based on the semantic connotation of their respective attributes.
Each image is paired with captions that describe the attributes as well as example user prompts.
Dataset Construction:
A total of 30,000 facial images are sampled from the CelebA dataset following the processing pipeline of CelebA-HQ, with a high resolution of 1024×1024.