Multi-scale Landmark Localization Network for 3D Facial Point Clouds

Facial landmark localization on 3D point clouds has been a major concern in the field of computer vision. Recent methods do not feature data containing multiple faces with large-scale variance, which has become increasingly common with the rapid development and wide application of 3D imaging technology. In this paper, we propose a Multi-scale Landmark Localization network for 3D facial point clouds.We evaluate the proposed method on the dataset synthesized by appending and scaling the data in the public dataset BU3DFE to demonstrate the robustness and efficiency. Upon comparing the proposed method with other methods on the standard dataset BU3DFE, in which data only contain one face with smallscale variance, we find that the proposed method shows higher or comparable performance with mean localization errors of 3.34 2.19 mm.