Unsupervised Facial Landmark Detection

The unsupervised facial landmark detection task aims to learn embedded representations of facial features through unlabeled data and then use these embeddings to train simple regressors to predict the locations of facial landmarks. This method first learns low-dimensional embeddings of images in an unsupervised environment, and subsequently uses a regression model to recover the coordinates of the landmarks from these embeddings, thereby achieving efficient and accurate facial landmark localization. This technique has significant application value in the field of computer vision, especially in reducing annotation costs and improving model generalization on large-scale datasets.