How did you train GrabNet on OakInk? and is this code available?

[AmitaiOvadia]

Quoting from your paper:

"In this work, we use GrabNet [68], trained on object shapes from OakInk [83], to generate hand poses as reward guidance for our policy training."

Did you use the GrabNet code version from five years ago? I ask because it is currently difficult to use due to numerous bugs.
Also, do you have the trained GrabNet models available? If so, could you share the code for evaluating them on OakInk?

Thank you!
Amitai