UIRNet: Facial Landmarks Detection Model with Symmetric Encoder-Decoder

초록

One of the challenging problems for facial landmarks detection is learning important features from faces that contain different deformation of face shapes and pose. These important features include eye centres, jawline points, nose points, mouth corners etc that are helpful in various computer vision-related applications. The detection of facial landmarks is difficult when faces have a lot of variation in different conditions. These conditions could be various imaging conditions such as illumination, occlusion, or head poses. In this paper, we propose a deep learning-based facial landmarks detection model called Unet-Inception-ResNet (UIRNet) to predict distinct feature points. The model predicts 68-point landmarks from the detected faces from digital images or video. © 2022 IEEE.

키워드