Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add support mip nerf in training example #129

Open
wants to merge 5 commits into
base: master
Choose a base branch
from

Conversation

foreverYoungGitHub
Copy link

Hi, @liruilong940607,

Thanks for your awesome work!

I just add the support for training the mip nerf (kind of) with nerfacc based on your discussion at #88. I did a quick test for the lego scene with 1 t4 and this is the full logs:

elapsed_time=1.76s | step=0 | loss=0.07456 | alive_ray_mask=64 | n_rendering_samples=10341 | num_rays=64 |
elapsed_time=675.89s | step=5000 | loss=0.00614 | alive_ray_mask=1972 | n_rendering_samples=65941 | num_rays=6340 |
elapsed_time=1436.05s | step=10000 | loss=0.00325 | alive_ray_mask=2624 | n_rendering_samples=62479 | num_rays=8388 |
elapsed_time=2252.93s | step=15000 | loss=0.00259 | alive_ray_mask=3201 | n_rendering_samples=65802 | num_rays=10022 |
elapsed_time=3095.50s | step=20000 | loss=0.00191 | alive_ray_mask=3379 | n_rendering_samples=65854 | num_rays=10483 |
elapsed_time=3944.66s | step=25000 | loss=0.00169 | alive_ray_mask=3569 | n_rendering_samples=66453 | num_rays=11441 |
elapsed_time=4818.70s | step=30000 | loss=0.00130 | alive_ray_mask=3888 | n_rendering_samples=65518 | num_rays=12459 |
elapsed_time=5703.86s | step=35000 | loss=0.00117 | alive_ray_mask=4004 | n_rendering_samples=65890 | num_rays=12501 |
elapsed_time=6595.80s | step=40000 | loss=0.00116 | alive_ray_mask=4163 | n_rendering_samples=66805 | num_rays=13105 |
elapsed_time=7495.55s | step=45000 | loss=0.00095 | alive_ray_mask=4178 | n_rendering_samples=66799 | num_rays=13154 |
elapsed_time=8396.90s | step=50000 | loss=0.00086 | alive_ray_mask=4242 | n_rendering_samples=67710 | num_rays=13257 |
100%|██████████| 200/200 [19:52<00:00,  5.96s/it]
evaluation: psnr_avg=33.62006133079529

The training time is pretty similar with vanilla nerf and the performance is almost same as well (which I do expect slightly better performance).

It would be nice if you can take a look at it when you are available!

Thank you in advance!

Btw, this pr also changes the dataset root dir to an argument for easier usage 😉

@foreverYoungGitHub
Copy link
Author

@liruilong940607 @tancik I'm wondering if you can take a look at it when you are available?

@liruilong940607
Copy link
Collaborator

Hey thanks for the implementation! I quickly skimmed it and it looks all reasonable! I will take a close look in the weekend.

@foreverYoungGitHub
Copy link
Author

foreverYoungGitHub commented Dec 9, 2022

That would be super helpful! Thank you Li!

parser.add_argument(
"--data_root",
type=str,
default=str(pathlib.Path.home() / "data"),
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

you might miss importing the pathlib at the beginning

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants