【论文阅读笔记】HyperFace: A Deep Multi-task Learning Framework for Face Detection, Landmark Localization

《HyperFace: A Deep Multi-task Learning Framework for Face Detection, Landmark Localization, Pose Estimation, and Gender Recognition》

Ranjan R, Patel V M, Chellappa R. Hyperface: A deep multi-task learning framework for face detection, landmark localization, pose estimation, and gender recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 41(1): 121-135.

多任务网络：face detction, landmarks localization , lanmarks visibility , face pose, gender recongnition

使用selective search theory产生proposal region

网络结构一：基于alexnet

网络结构二：基于resnet

两种post processing 策略：改进传统非极大值抑制（NMS)，使识别结果更加鲁棒。

IRP:

L-NMS:

 

使用 single shot detector(SSD) 代替Selective search ,实现fast-HyperFace

Selective search 是R_CNN中使用的region proposal technology。