M2FPA

M2FPA数据集是由中国科学院自动化研究所模式识别国家重点实验室创建的大规模多视角人脸数据库，包含397,544张图像，涵盖229个对象，涉及多种姿态、属性、光照和配件变化。该数据集通过专门设计的多相机采集系统捕获，确保了图像的高分辨率和姿态的精确多样性。M2FPA数据集旨在推动面部姿态分析的研究，特别是在面部正面化、面部旋转、面部姿态估计和姿态不变面部识别等领域。The M2FPA dataset is a large-scale multi-view face database developed by the State Key Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences. It contains 397,544 images covering 229 subjects, with variations in multiple dimensions including pose, attributes, illumination and accessories. Captured via a specially designed multi-camera acquisition system, the dataset guarantees high image resolution and precise diversity of facial poses. The M2FPA dataset aims to advance research on facial pose analysis, especially in the fields of face frontalization, face rotation, facial pose estimation and pose-invariant face recognition.

搜集汇总

背景与挑战

背景概述

M2FPA是一个由中国科学院自动化研究所等机构构建的大规模面部姿态分析数据集，包含397,544张图像，涵盖229个对象、62种姿态（包括偏航和俯仰角变化）、4个属性和7种光照条件，适用于学术研究。数据集提供裁剪图像、注释和测试协议，主要用于计算机视觉和模式识别领域的面部姿态分析任务。