UV Volumes for Real-time Rendering of Editable Free-view Human Performance

an Jiaotong University 3 Ant Group 4 Tencent AI Lab (c) Retexturing (e) Reposing (d) Reshaping (a) Novel View Synthesis DyNeRF NeuralBody Animatable-NeRF Ours 1.5FPS 1.3FPS 0.3FPS 42FPS (b) UV Avatar Figure 1. We decompose a dynamic human into 3D UV Volumes along with a 2D texture. The disentanglement of appearance and geometry enables us to achieve (a) high-fidelity real-time novel view synthesis guided by (b) a smooth UV avatar, (c) retexturing of a 3D human by editing a 2D texture, (d) reshaping and (e) reposing by changing the parameters of a human model while keeping the texture untouched.