TinyFace
收藏资源简介:
TinyFace 是一个大规模的人脸识别基准,有助于在深度学习中以大规模(大型画廊人口规模)研究本地 LRFR(低分辨率人脸识别)。 TinyFace 数据集由 5,139 个标记的面部身份组成,这些身份由 169,403 个为 1:N 识别测试设计的原生 LR 人脸图像(平均 20×16 像素)给出。 TinyFace 中的所有 LR 人脸都是从各种成像场景的公共网络数据中收集的,这些数据是在姿势、光照、遮挡和背景的不受控制的观察条件下捕获的。
TinyFace is a large-scale face recognition benchmark that facilitates research on local low-resolution face recognition (LRFR) in deep learning with large gallery population sizes. The TinyFace dataset consists of 5,139 labeled facial identities, represented by 169,403 native low-resolution (LR) face images designed for 1:N identification tests, with an average resolution of 20×16 pixels. All LR faces in TinyFace are collected from public web data across various imaging scenarios, captured under uncontrolled observation conditions including pose variations, illumination changes, occlusions and background variations.




