【DeepFace】《DeepFace：Closing the Gap to Human-Level Performance in Face Verification》

CVPR-2014

Taigman Y, Yang M, Ranzato M A, et al. Deepface: Closing the gap to human-level performance in face verification[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2014: 1701-1708.

1、Background and Motivation

当前人脸验证算法流程如下

detect ⇒ align ⇒ represent ⇒ classify

性能与人类视觉系统之间仍存在差距。

作者基于 3D 人脸对齐和 DNN 分类方法，提出 DeepFace 人脸验证流程，效果逼近人类水平

2、Related Work

• Big data and deep learning
• Face recognition state of the art
  border-control and smart biometric identification，hand-crafted features，tens of thousands of image descriptors

3、Advantages / Contributions

• 提出了 DeepFace 人脸验证方法（couples large feedforward-based models with fine 3D alignment.）

• LFW 数据集上逼近人类水平，YTF 数据集上错误率降低了 50% 以上

4、Method

4.1、Face Alignment

人脸对齐的作用，by providing a normalized input

we describe a system that includes analytical 3D modeling of the face based on fiducial points, that is used to warp a detected facial crop to a 3D frontal mode (frontalization).

关键点检测->2D Alignment->3D Alignment->Frontalization

这部分细节比较多，感兴趣可以对着公式表达推导下

4.2、Representation

descriptor engineering vs through a large