Question about Telegathering

[junhua-l]

Hi, thank for your excellent work! I am a HCI researcher. I noticed that NeuMan supports Tele-gathering. I want to build a real-time capture streaming and visualize the digital human for multi-user social experience. Could Novel view synthesis be rendered in real-time? (what's the fps of novel view synthesis?).

My understanding is that we get the 3D model after human NeRF reconstruction and Scene NeRF reconstructions. Then for the rest frame, we could update the 3D model given only the pose information (smpl parameters) and then synthesize novel view. Does the process need re-training?