-
Notifications
You must be signed in to change notification settings - Fork 190
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
Implementation of BakedSDF #50
Comments
+1 would love to see this implementation. The results of BasedSDF are really amazing. |
Hi, BakedSDF is implemented. After downloading the mipnerf360 dataset with
or
Here are some results: Tuning some hyper parameters would likely give better results. |
Wow, you got that implemented extremely fast! Excellent, I'll give it a test! |
To what extent is |
Looking at the code, unless I'm missing something, the quantization-aware training is not included either. Is this the case or am I missing something? |
@niujinshuchong can you advise & @pablovela5620 did you try it out? |
@reynoldscem Training of bakedsdf and mesh extraction are supported. You could extract the mesh after training with ns-extract-mesh --load-config CONFIG --output-path MESHFILE.ply --bounding-box-min -2.0 -2.0 -2.0 --bounding-box-max 2.0 2.0 2.0 --resolution 512 --marching_cube_threshold 0.001 --create_visibility_mask True The baking part is missing and I don't think it will be added soon. Contributions are welcome. |
Thanks for your great work, I try to train bakedSDF in garden scene of nerfstudio-data-mipnerf360 data with following cmd: ns-train bakedsdf --vis viewer --experiment-name garden-bakedsdf then extrash mesh: ns-extract-mesh --load-config outputs/garden-bakedsdf/bakedsdf/2023-03-07_110420/config.yml The mesh like this: The surrounding mesh is incomplete, and I adjust the bounding-box and resolution, the result doesn't improved, please help me where is the problem. |
@tianxiaguixin002 I think this might because we compute the visibility grid with testing images. I push a update for that. Here is the results using all training images to compute the visibility grids. MLP's mesh is more smooth. |
Thank you very much for your reply, I will try again in a few days. |
how can I convert the result to a normal glb file, which can be rendered in threejs? |
@tlightsky as far as I can tell, that part isn't included |
@tlightsky @reynoldscem Yes, It's not included. |
@niujinshuchong I use the latest code and have trained bakedsdf-mlp in mipnerf360-garden scene with the offical cmds from sdfstudio-examples.md. The mesh quality is good under normal observation position. like following picture:
|
@tianxiaguixin002 seems okay, you can download the official bakedsdf .glb file from https://dl.dropboxusercontent.com/s/7srlfofbiryehn6/gardenvase.glb and compare with this. I think they are very similar. the far region distortion you see is common, a bbox crop is required to clear those. |
@niujinshuchong What exactly is the difference between |
@gaoalexander |
Hi, I use
can anybody give me some advice? |
@CanCanZeng Could you try to use |
Thank you for your advice @niujinshuchong , it works! |
Hi, how long does bakedsdf to be trained to have a good result? |
https://bakedsdf.github.io/, most of the components are already there (VolSDF | MipNerf 360 | Marching cubes). The largest missing part is the
appearance model that uses spherical Gaussians
. Would love to see this brought into SDFStudioThe text was updated successfully, but these errors were encountered: