[Question] Looking to use another dataset to build a point cloud?

[pragatss]

I am looking to use a different dataset to build a point cloud using https://github.com/leggedrobotics/viplanner/blob/main/viplanner/depth_reconstruct.py.

I've got my folder directly setup as stated:

    - env_name
        - camera_extrinsic.txt  (format: x y z qx qy qz qw)
        - intrinsics.txt (expects ROS CameraInfo format --> P-Matrix)
        - depth  (either png and/ or npy)
            - xxxx.png  (images should be named with 4 digits, e.g. 0000.png, 0001.png, etc.)
            - xxxx.npy  (arrays should be named with 4 digits, e.g. 0000.npy, 0001.npy, etc.)
        - semantics (optional)
            - xxxx.png  (images should be named with 4 digits, e.g. 0000.png, 0001.png, etc., RGB images)

and the script begins to run but fails with

updating open3d geometry point cloud with last images ...
Segmentation fault (core dumped)I

Ultimately I would like to train with depth, segmentation, intrinsics and extrincs from a different dataset.

Are there any guidelines or specifications on how to do that? Any idea if this is a data problem or a cpu problem. If it is data, then is there a specific format the input data need to be. Is there any example depth/segmentation data readily available for me to compare. I wonder if this may have something to with the channels in my images.

Please let me know if you have any more questions.

[GalaxyPatrick]

You can downgrade the version of the numpy library which above 2.0 causes this problem. isl-org/Open3D#6840 (comment)