Masked Face-Landmark Prediction with Mask-Coefficient

In the past, face-landmark prediction works, less optimization was forced on the non-masked face, and most focused on the full face. In this paper, we propose a method to improve the effectiveness of neural network training through Mask-Coefficient based on the proportion of the masked face. We predict facial feature points by training a Regular Network model. During the training process, the penalty value of the loss function is adjusted according to the percentage of the face being occluded, thereby enhancing the model's ability to predict occluded facial images. Finally, we demonstrate the effectiveness of the proposed method through several metrics under different mask ratios and publish several practical examples for verification. The experimental results of inter-eye normalization (ION) and inter-pupillary normalization (IPN) measurements for face feature point prediction are better than previous methods.