Adaptive robust loss for landmark detection

Yingjie Tian, Duo Su, Shilin Li

Abstract: Highlights•The properties and comparisons of the landmark detection losses are presented.•A novel unified Adaptive Robust loss for landmark detection task is proposed.•A lightweight multi-scale fusion network is built based on MobileNetV2.•Extensive experiments on face alignment and human pose estimation are carried out.•Some frontier challenges and research directions are also discussed in our article.