Modified Stacked Hourglass Networks for Facial Landmarks Detection.

Facial landmarks detection is a fundamental research topic in computer vision. This topic has been largely improved recently thanks to the development of convolution neural networks (CNN). This paper proposes a modified version of the Stacked Hourglass Network, which is a state-of-the-art architecture for landmark localization. Instead of using the original residual block, this paper uses the \(\lambda \)-residual-block to get more effective features. The proposed network can achieve better result than other state-of-the-art methods on two very challenging 3D facial landmark datasets, Menpo-3D and 300 W.