search

DavidXu-JJ / DebSDF

[TPAMI 2024] DebSDF: Delving into the Details and Bias of Neural Indoor Scene Reconstruction
https://davidxu-jj.github.io/pubs/DebSDF/
MIT License
101 stars 0 forks source link

the evaluation resolution of t&t dataset #8

Closed dawnzyt closed 19 hours ago

dawnzyt commented 1 day ago

Hello, thanks for opening this great work. I have a question about the evaluation resolution of t&t dataset when submitting to the t&t official website. I find that 'final_mesh_resolution' is set to 1024 in 'tnt_grids_x.conf' in your code, if this means the quantitative results on t&t datasets presented in your paper are all evaluated at 1024 resolution? I knew that monosdf is evaluated at a resolution of 768, is debsdf consistent with monosdf?

DavidXu-JJ commented 23 hours ago

Hi Ziyu,

Thanks for your attention to this work.

The code of this project is adapted from the resolution of t&t at MonoSDF's config:

https://github.com/autonomousvision/monosdf/blob/12513009cd20a08b35ddc2c55a7e42da2c7baf43/code/confs/tnt_grids_1.conf#L15

I'm personally not 100% sure whether the resolution in MonoSDF's config is aligned with its paper's experiment or their submitted mesh file.

But our results in the DebSDF paper are evaluated at 1024 resolution as the config mentioned above.

Since it has been a while, I may not be able to find the exact checkpoint of t&t we used for mesh evaluation.

If you're certain that MonoSDF is 768 resolution for their actual paper's evaluation, you may wish to re-run DebSDF training or MonoSDF training and extract their meshes at the same resolution you set.

Thanks for your question.

Best, Jingwei

DavidXu-JJ commented 19 hours ago

Feel free to reopen this issue if necessary. Or just comment under this issue.

dawnzyt commented 19 hours ago

Feel free to reopen this issue if necessary. Or just comment under this issue.

Ok, thank you for your reply. this helps me a lot! by the way, I learned from the monosdf issue that the resolution is set to 768: https://github.com/autonomousvision/monosdf/issues/2#issuecomment-1184108254

DavidXu-JJ commented 19 hours ago

Thanks for your notification! We may have accidentally missed this part. Sorry about that. Thank you very much for letting us know. That also helps me a lot.