LiveHand: Real-time and Photorealistic Neural Hand Rendering

The human hand is the main medium through which we interact with our surroundings. Hence, its digitization is of uttermost importance, with direct applications in VR/AR, gaming, and media production amongst other areas. While there are several works for modeling the geometry and articulations of hands, little attention has been dedicated to capturing photo-realistic appearance. In addition, for applications in extended reality and gaming, real-time rendering is critical. In this work, we present the first neural-implicit approach to photo-realistically render hands in real-time. This is a challenging problem as hands are textured and undergo strong articulations with various pose-dependent effects. However, we show that this can be achieved through our carefully designed method. This includes training on a low-resolution rendering of a neural radiance field, together with a 3D-consistent super-resolution module and mesh-guided space canonicalization and sampling. In addition, we show the novel application of a perceptual loss on the image space is critical for achieving photorealism. We show rendering results for several identities, and demonstrate that our method captures pose- and view-dependent appearance effects. We also show a live demo of our method where we photo-realistically render the human hand in real-time for the first time in literature. We ablate all our design choices and show that our design optimizes for both photorealism and rendering speed. Our code will be released to encourage further research in this area.