





📢📢📢 Neural Inverse Rendering from Propagating Light 💡 Our CVPR Oral introduces the first method for multiview neural inverse rendering from videos of propagating light, unlocking applications such as relighting light propagation videos, geometry estimation, or light component separation! \w @imarhombus (co-first), @AndrewEJXie, @mpotoole, and @DaveLindell

Joint work with Dorain Chan, @mpotoole, and Srinivasa Narasimhan Questions? Comments? Ideas? Come say hi at the Friday morning poster session! (poster 32a) Paper: tinyurl.com/2p9ap873 4/4

Here are some highlights in video form! Our method is robust to C-ToF phase-wrapping, low-reflectance noise, multi-path interference, and erroneous camera calibration.

TöRF: Time-of-Flight Radiance Fields for Dynamic Scene View Synthesis pdf: arxiv.org/pdf/2109.15271.pdf abs: arxiv.org/abs/2109.15271 project page: imaging.cs.cmu.edu/torf/