S3DIS image inputs and visibility files are referenced but preprocessing pipeline is missing

[nemanjajoe]

Hi, thanks for the great project.

I noticed that the S3DIS image-based model refers to the following data paths:

• data/s3dis_images/{Area}/{Room}/color/
• data/s3dis_visibility/{Area}/{Room}/visibility/

These files are consumed by pointcept/datasets/s3dis.py through:

• *_mask.npy
• *_points.npy
• *.png

However, in the current repository I could not find a preprocessing or generation script that produces these two directories for S3DIS.

What I found:

• pointcept/datasets/preprocessing/s3dis/preprocess_s3dis.py only generates the 3D room-level files such as coord.npy, color.npy, segment.npy, and instance.npy
• I could not find a script in the repo that generates:
  • data/s3dis_images/...
  • data/s3dis_visibility/...

Could you please clarify:

1. How are s3dis_images and s3dis_visibility generated?
2. Are they derived from Stanford 2D-3D / a separate preprocessing pipeline?
3. Is there any script, dataset source, or documentation that should be added to the repository?

If possible, it would be very helpful to include:

• the exact preprocessing steps,
• required raw data sources,
• and a script to generate these files reproducibly.

Thanks!

[YilmazKadir]

Hi, thanks for pointing this out. The preprocessing script for generating the image-based S3DIS inputs is missing. I will clean up the script and push it shortly.

[YilmazKadir]

I uploaded the dataset preprocessing instructions for S3DIS. Please let me know if you encounter any problems.