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

Not running for "all55" version on "interpolate_posterior.sh" #66

Open
dqj5182 opened this issue Feb 23, 2024 · 4 comments
Open

Not running for "all55" version on "interpolate_posterior.sh" #66

dqj5182 opened this issue Feb 23, 2024 · 4 comments

Comments

@dqj5182
Copy link

dqj5182 commented Feb 23, 2024

I wish to do interpolation with "all55" model but the "interpolate_posterior.sh" does not work with the following error.
Could you give insight on how to solve this issue?

Screen Shot 2024-02-23 at 9 00 18 PM

@noahcao
Copy link

noahcao commented Mar 26, 2024

@ZENGXH I am facing the same problem here. what is the definition of the missing file of datasets.neuralspline_datasets? Also, does this mean that unconditional/all55 checkpoint is trained by a different dataset convention from others?

@ZENGXH
Copy link
Collaborator

ZENGXH commented Mar 26, 2024

@dqj5182 you could set data.type to be datasets.pointflow_datasets; if you want to do interpolation of all55 class, you need to load the all55 dataset, but the training dataset is not released. one thing you could do is to modify the dataloader a bit to load the released validation point cloud for the all55 model.

Note the coordinates system of the all55 data are a bit different from the released pointflow datasets, so if you want use pointflow point cloud for poterior interpolation you need to write a small function to convert the coordinates system, you could use the validation point cloud for the all55 model as reference to check if that's correct or not

@ZENGXH
Copy link
Collaborator

ZENGXH commented Mar 26, 2024

@noahcao the two dataset are all from shapenet, the difference is all55 has different coordinates system, and all shape are using box normalization.
the all55 shapes are converted from the surface so there is not points indicating the internal structure

@ZENGXH
Copy link
Collaborator

ZENGXH commented Mar 26, 2024

example scripts to convert the coordinates: https://github.com/nv-tlabs/LION/blob/main/utils/render_mitsuba_pc.py#L49-L50
I use this when convert the all55 dataset (we called it Shapenet-vol in the paper) points to be in the same coordinates as the point flow (we called it Shapenet dataset in the paper). (for the use case here, we want to convert pointflow pointcloud back to the corrdinate system of all55 dataset)

it's always better to visualize them to make sure the data is align with the validation point cloud of all55 model to make sure you get correct input for the all55 model.

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

No branches or pull requests

3 participants