CelebA-Dialog

CelebA-Dialog是具有以下功能的大型视觉语言人脸数据集。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，都有描述属性的文本标题和用户编辑请求示例。 CelebA-Dialog具有: 10,177数量的身份，202,599数量的面部图像和每个图像的5个细粒度属性注释: 刘海，眼镜，胡须，微笑和年龄，文本标题和每个图像的用户编辑请求。CelebA-Dialog is a large-scale vision-language facial dataset with the following features. Its facial images are equipped with rich fine-grained labels, which are divided into multiple levels based on the semantic meaning of each corresponding attribute. Along with each image, there are text captions describing the attributes and examples of user editing requests. CelebA-Dialog includes 10,177 unique identities, 202,599 facial images, 5 fine-grained attribute annotations per image covering bangs, eyeglasses, beard, smile and age, as well as text captions and user editing requests for each image.

搜集汇总

背景与挑战

背景概述

CelebA-Dialog是一个大规模视觉语言人脸数据集，包含10,177个身份和202,599张面部图像，每张图像都标注了5个细粒度属性（如刘海、微笑）、文本标题和用户编辑请求。该数据集由南洋理工大学实验室和香港中文大学于2021年发布，旨在支持面部属性分析和编辑任务。