InstaFace: Single-View 3D Face Reconstruction with Hair Try-On

This study presents InstaFace, an innovative system designed for the reconstruction of 3D human heads from single images, enabling the distinct reconstruction of both the 3D head model and the 3D hair model. In contrast to previous research that represented hair as a mesh with attached textures, InstaFace introduces a novel approach by generating the hair component independently, achieving remarkable high-detail accuracy and offering a range of customizable features, particularly suited for applications such as hairstyle modification. The modular separation of the reconstruction process within InstaFace allows for the seamless integration of different reconstruction methods. In order to do that, we introduce an alignment algorithm that seamlessly combines facial reconstruction modules (DECA \(+\) MICA) and the specialized hair reconstruction (HairNet-based). The evaluations of InstaFace reveal promising results when compared to similar models, as demonstrated by favorable PSNR, SSIM, and LPIPS scores, underscoring the effectiveness of this proposed approach in the realm of 3D head reconstruction.