search

nianticlabs / diffusionerf

[CVPR 2023] DiffusioNeRF: Regularizing Neural Radiance Fields with Denoising Diffusion Models
MIT License
286 stars 15 forks source link

About the method of geometry reconstruction #10

Open xiaobiaodu opened 9 months ago

xiaobiaodu commented 9 months ago

Hi

Thanks for your excellent work. I am very interested in Fig. 6. Which method you used to reconstruct the geometry? If you can provide a GitHub repo, I would appreciate a lot.

image

jamiewynn commented 8 months ago

Hi there,

To reconstruct the geometry we extracted an isosurface from the nerf with marching cubes (see extract_geometry in nerf/utils.py), and then additionally culled it using this code from MonoSDF: https://github.com/autonomousvision/monosdf/blob/main/dtu_eval/evaluate_single_scene.py

Best, Jamie Wynn

ZepSbosnia commented 1 week ago

Hi, I'm also interested in this issue. I would like to have more information about how to exploit evaluate_single_scene.py from MonoSDF. Indeed running it over the mesh obtained from NeRF does not lead to any improvements on the mesh. Additionally which mesh resolution do you have provide to the marching cubes method ?