CelebA-HQ高质量人脸图像数据集

CelebAHQ数据集主要面向人脸生成、隐私保护与模式识别等相关研究需求，在国家重点研发计划（2023YFE0202700）支持下，基于高质量名人人脸属性数据的研究背景建设而成，以 CelebA-HQ 为数据来源基础，通过统一的预处理与匿名化、可逆恢复方法生成，系统记录了原始人脸、匿名化人脸及恢复人脸三类数据，完整覆盖身份、属性保持与隐私保护等关键研究内容，数据集共包含三个子文件夹（ori/anony/recover），每类各 30000 张 128×128 分辨率的彩色人脸图像，并配套属性与结构一致性信息，总体规模约 313.96MB，采用 JPG 格式、完全共享方式发布，在高质量人脸匿名化评测、可逆隐私保护与生成模型研究中具有重要的实验与基准意义。

The CelebAHQ dataset is primarily designed for research related to face generation, privacy protection, pattern recognition and other relevant fields. Supported by the National Key R&D Program of China (Grant No. 2023YFE0202700), it was constructed based on the research background of high-quality celebrity face attribute data. Built on the CelebA-HQ dataset as the source data, it is generated through unified preprocessing, anonymization and reversible recovery methods. It systematically records three types of data: original faces, anonymized faces and recovered faces, fully covering key research contents such as identity preservation, attribute retention and privacy protection. The dataset consists of three subfolders (ori/anony/recover), each containing 30,000 128×128 resolution color face images, and is accompanied by attribute and structural consistency information. The total size of the dataset is approximately 313.96 MB, and it is released in JPG format with full open access. It holds significant experimental and benchmark value for high-quality face anonymization evaluation, reversible privacy protection and generative model research.

搜集汇总

背景与挑战

背景概述

CelebA-HQ高质量人脸图像数据集是一个面向人脸生成、隐私保护与模式识别研究的高质量数据集，包含原始、匿名化和恢复三类人脸图像，每类30000张128×128分辨率的彩色图像，总体规模约313.96MB。该数据集采用JPG格式，配套属性与结构一致性信息，适用于高质量人脸匿名化评测、可逆隐私保护与生成模型研究。