search

jkulhanek / tetra-nerf

Official implementation for Tetra-NeRF paper - NeRF represented as triangulation of input point cloud.
https://jkulhanek.com/tetra-nerf
MIT License
278 stars 14 forks source link

Does the "ns-train tetranerf colmap " work? #29

Open samcao0416 opened 9 months ago

samcao0416 commented 9 months ago

Hi, thank you for your work.

I followed your instruction to run mip nerf 360 data successfully with four commands. When I was trying to run my own data with ns-train tetranerf colmap --data, an error occurred called File "/media/stereye/f1a736d6-e532-4ef2-a3b9-493fb9fb7ab/sam/codes/nerfstudio/nerfstudio/cameras/cameras.py",line 288,in is jagged h jagged = not torch.all(self.height == self.height.view(-1)[0]) RuntimeError: CUDA error:device-side assert triggered CuDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA LAUNCH BLOCKING=1 Compile with TORCH USE CUDA DSA' to enable device-side assertions.

I also tried ns-train tetranerf colmap with mipnnerf 360 dataset, since it also accord with the colmap data structure. The same errors occurred. So, does the command work? image

jkulhanek commented 2 months ago

Hi, it should work. The issue is that nerfstudio rounds downscaled image sizes differently than Barron did in the MipNeRF 360 dataset. You have to manually change the rounding. I assume you get a similar error when trying to run any other method on the dataset. In this repository, we provide a custom dataloader to fix the issue.

cmh1027 commented 1 month ago

For anyone stuck on this problem, go to nerfstudio/data/dataparsers/colmap_dataparser and change the default value of "downscale_rounding_mode" to "round" (or "ceil", whatever that does not trigger an error)