MegaFace

MegaFace数据集由华盛顿大学计算机科学与工程系创建，包含超过100万张照片，涵盖69万不同个体。该数据集旨在评估大规模人脸识别算法的性能，特别是在面对大量干扰项（从10到100万）时的表现。数据集内容丰富，包括各种姿态、表情、光照和曝光条件下的照片，主要来源于Yahoo的Flickr照片库。创建过程中，通过分析Flickr用户ID和群组照片，优化了独特身份的数量。MegaFace数据集的应用领域广泛，主要用于推动和评估大规模人脸识别技术的发展，特别是在需要高精度识别大量不同个体的场景中。

The MegaFace dataset was developed by the Department of Computer Science & Engineering at the University of Washington. It comprises over one million photographs encompassing 690,000 distinct individuals. This dataset is intended to evaluate the performance of large-scale face recognition algorithms, particularly their efficacy when confronted with a large number of distractors ranging from 10 to 1,000,000. Featuring diverse content, the dataset includes photos captured under varying poses, expressions, lighting conditions and exposure settings, and is primarily sourced from the Yahoo Flickr photo library. During its construction, the number of unique identities was optimized by analyzing Flickr user IDs and group photographs. The MegaFace dataset has a wide range of application scenarios, and is mainly used to advance and assess the development of large-scale face recognition technologies, especially in scenarios requiring high-precision identification of a large number of distinct individuals.

搜集汇总

背景与挑战

背景概述

MegaFace是一个大规模人脸识别数据集，包含100多万张照片和69万不同个体，用于评估算法在大量干扰项下的性能。数据集涵盖多种姿态、表情和光照条件，主要来自Flickr照片库，适用于高精度大规模人脸识别场景。