Zhaoxi Chen
Zhaoxi Chen
Close due to inactivity. Feel free to reopen for further issues!
Updated!
Close due to inactivity. Feel free to reopen for further issues!
@mikedh Hey! I've updated the tests accordingly with all run tests passed.
As specified in our supplementary doc, we train the denoiser on 4 A100 GPUs with a batch size of 8 on each GPU for 200,000 iterations.
To relight a static avatar with a rotating envmap as background, you can simply take the first frame (typically any single frame) when relighting a dynamic sequence, which will freeze...
We are using the PrimX as our 3D representation instead of 3DGS or NeRF. For convertion, you need to render the output into multiview images and do reconstruction.