A complementary regression network for accurate face alignment
作者：
Highlights：
• We propose CRN that combines global and local regression networks for face alignment.
• CRN converts GRN's coordinates to LRN's heatmap, vice versa, to get landmark points.
• CRN works complementarily that GRN and LRN compensate their demerits each other.
• We conducted several experiments on 300-W public/private dataset Menpo dataset.
• We achieved 3.14%, 3.74%, 2% SOTA face alignment accuracy in terms of PNME.
摘要
•We propose CRN that combines global and local regression networks for face alignment.•CRN converts GRN's coordinates to LRN's heatmap, vice versa, to get landmark points.•CRN works complementarily that GRN and LRN compensate their demerits each other.•We conducted several experiments on 300-W public/private dataset Menpo dataset.•We achieved 3.14%, 3.74%, 2% SOTA face alignment accuracy in terms of PNME.
论文关键词：Facial landmark detection,Complementary regression network,Coordinate-to-heatmap transform,Heatmap-to-coordinate transform
论文评审过程：Received 29 July 2019, Revised 18 December 2019, Accepted 6 January 2020, Available online 22 January 2020, Version of Record 3 February 2020.
论文官网地址：https://doi.org/10.1016/j.imavis.2020.103883
