search

kwea123 / ngp_pl

Instant-ngp in pytorch+cuda trained with pytorch-lightning (high quality with high speed, with only few lines of legible code)
MIT License
1.26k stars 156 forks source link

Synthetic NeRF Dataset creation tool #17

Closed cubantonystark closed 2 years ago

cubantonystark commented 2 years ago

Not really an issue but just a small PSA to contribute to your effort, if allowed. I just uploaded a Synthetic NeRF dataset creation tool here. Please feel free to critique and enhance and if deemed usable incorporate onto your effort. I am amazed at your work and felt the need to humbly contribute . I'm working on a GUI as well. Regards

kwea123 commented 2 years ago

Thank you!!

cubantonystark commented 2 years ago

Hey @kwea123 !

My pleasure! Also, thanks for pinning to the top of the issue list for visibility. It is great being able to help somehow.

Question: how feasible would it be to create an mesh (.obj or ply) from the checkpoints with the old colab notebooks for nerf_pl? I'm really interested in that part. I also took out some of he hodgepodge code in the dataset generator to make it more compatible with ngp_pl. I'm by no means an expert as you are and hope you and the community at large are able to use my code as well.

Regarding using the colab notebooks; some advise would be appreciated.

kwea123 commented 2 years ago

To create a mesh, the process is the same:

  1. generate dense grids in [-0.5, 0.5]^3 (you can use kornia create_meshgrid3d)
  2. query the model for the density at these points (use model.density())
  3. finally use marching cubes

I've successfully done this on my side, but the code is messy, I will upload this later

cubantonystark commented 2 years ago

I’ll jump on that and see what i come up with. Thank you!