Datasets:

Languages:
English
ArXiv:
License:

preprocessing suggestions

#9
by SFXX - opened

Hi Matt @mattdeitke
Thanks for the excellent contribution of Objaverse!

I'm trying to do some preprocessing for this dataset recently. I noticed that the tutorial for the preprocessing part is not released yet, so any suggestions will be appreciated!

I mainly want to do two things:

  1. Render images
  2. extract point cloud with different numbers of points

I have done things like rendering ShapeNET to images using blender with the Stanford shapnet rendering scripts, but I find that it will require .obj file type. I found the Objaverse downloaded type is .glb, so that script might not be directly applicable here. Also, I'm not sure if there is a convenient tool to convert large-scale .glb files to .obj files efficiently via command line, will appreciate it if you happen to know any tool I can leverage!
I'm also not sure whether the rendering hyper-parameters for ShapeNET will work for Objaverse.

Thanks!

Hi @SFXX ! Thanks for the kind words :)

You do not need to do any conversion from glb to obj, Blender provides great support for working with GLBs! I've sent in a request to the legal team at AI2 to open source my Objaverse rendering scripts, so that should be public in the next day or so :)

Regarding extracting point clouds, I would recommend checking out the Point-E Blender rendering script. They extract point clouds :)

Hi @mattdeitke , thanks for the response! Will I be able to look for the script here "https://objaverse.allenai.org/docs/tools/blender" ?

SFXX changed discussion status to closed

Hi @SFXX , the scripts are here: https://github.com/allenai/objaverse-rendering

I'm still working on adding more documentation for them, but hope this help!

Sign up or log in to comment