Open Volleria opened 1 year ago
There is one thing that we do not support with nerfacc, which is to concat samples from coarse and fine sampling, as done in the original NeRF. However recent works kinda indicating this concatenation is not important.
Other than that, there is no logistical barrier I believe to use nerfacc for that codebase. Maybe I can better help you if you post what exactly is preventing you from using nerfacc in that repo.
I want to read recent works, just you said indicating original NeRF's concatenation is not important, but I didn't find them. Can you give some links? @liruilong940607
You can take a look at Mip-NeRF 360’s implementation: https://github.com/google-research/multinerf/blob/5d4c82831a9b94a87efada2eee6a993d530c4226/internal/models.py#L75
I have noticed that many nerf model codes that use nerfacc are different from nerf-pytorch. I am trying to use nerfacc in it but have not been successful yet, I would like to know if it is possible to use nerfacc in nerf-pytorch?