KEPLER: Simultaneous estimation of keypoints and 3D pose of unconstrained faces in a unified framework by learning efficient H-CNN regressors

作者：

Highlights：

• Present a cascade regression method for unconstrained face alignment

• Channeled Inception Net is presented for training in multi-task framework.

• Constrained training and local error correction significantly improve the performance.

• Impressive results on AFLW, AFW and COFW datasets

摘要

•Present a cascade regression method for unconstrained face alignment•Channeled Inception Net is presented for training in multi-task framework.•Constrained training and local error correction significantly improve the performance.•Impressive results on AFLW, AFW and COFW datasets

论文关键词：Face alignment,Keypoints,Landmarks,Deep networks,Convolution Neural Networks

论文评审过程：Received 2 October 2017, Revised 17 May 2018, Accepted 12 September 2018, Available online 25 September 2018, Version of Record 4 October 2018.

论文官网地址：https://doi.org/10.1016/j.imavis.2018.09.009