LFW（Labeled Faces in the Wild）人脸识别数据集

LFW数据集主要面向开放世界与无约束人脸识别、隐私保护及模式识别相关研究需求，在真实新闻媒体自然场景人脸数据（LFW）的研究背景下，依托国家重点研发计划（2023YFE0202700）建设形成，以原始 LFW 人脸图像为来源，通过统一的人脸预处理、匿名化生成与可逆恢复方法构建，系统记录了真实复杂环境下人脸的原始外观、匿名化结果及身份恢复结果等关键信息，数据集包含 ori、anony、recover 三个子文件夹，每类各 13,228 张 128×128 分辨率的彩色人脸图像，总体规模约 179MB，采用 JPG 格式并以完全共享方式发布，为复杂场景下人脸识别算法泛化能力评测、隐私保护技术验证及相关方法对比提供了具有代表性的标准化数据基础。

The LFW dataset is primarily developed to cater to research requirements related to open-world and unconstrained face recognition, privacy protection, and pattern recognition. Constructed under the National Key R&D Program of China (2023YFE0202700) against the research background of face datasets in natural real news media scenarios, it takes the original LFW face images as its source data. Built via unified face preprocessing, anonymization generation and reversible recovery methods, it systematically records key information including the original facial appearance, anonymized outputs and identity recovery results of faces in real complex environments. The dataset includes three subfolders: ori, anony and recover, each containing 13,228 color face images with a resolution of 128×128. The total size of the dataset is approximately 179 MB, with all images stored in JPG format and released under full open sharing. It provides a representative standardized data foundation for evaluating the generalization capability of face recognition algorithms in complex scenarios, verifying privacy protection technologies, and comparing related research methods.

搜集汇总

背景与挑战

背景概述

LFW人脸识别数据集是基于原始LFW人脸图像构建的标准化数据集，通过统一的人脸预处理、匿名化生成与可逆恢复方法，包含ori、anony、recover三个子文件夹，总计39,684张128×128分辨率彩色图像，总体规模约429.41MB。该数据集面向开放世界与无约束人脸识别、隐私保护及模式识别研究，为复杂场景下人脸识别算法泛化能力评测和隐私保护技术验证提供代表性数据基础，源自国家重点研发计划项目。