Exploring StyleGAN Latent Space for Face Alignment with Limited Training Data

With deep learning models growing in size over the years, sometimes exceeding a billion parameters now, the need for large, annotated training datasets grows too. To alleviate this problem, the interest in self-supervised learning is also increasing. In this domain, with the rise of Generative Adversarial Networks (GANs) and particularly StyleGAN, the quality of image generation is significantly improving. In this paper, we propose to use StyleGAN to perform face alignment with limited training data instead of image generation. Our proposed framework Face Alignment using StyleGAN Embeddings (FASE) projects real images into StyleGAN latent space and then predicts facial landmarks from the latent vectors. Our method achieves state-of-the-art on multiple face alignment datasets in the few-shot setting.