You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Thanks for your excellent work!
Here is a question about the data processing.
I found objects in Objaverse dataset are not in the same initial pose, especially the rotation angle.
I guess that's why your code have the args of rotaion_y and rotation_x.
How do you deal with the angle problem during training? Since it is impossible to set rotation manually during training.
Do you do some preprocess steps to make the rotations uniform?
The text was updated successfully, but these errors were encountered:
Hi, thanks for your interest. That's a good question.
We didn't have any preprocessing steps as we trained our lightcontrolnet only on random 2D renderings. Thus, there was no need for us to specify a canonical rotation for every object.
Hi,
Thanks for your excellent work!
Here is a question about the data processing.
I found objects in Objaverse dataset are not in the same initial pose, especially the rotation angle.
I guess that's why your code have the args of
rotaion_y
androtation_x
.How do you deal with the angle problem during training? Since it is impossible to set rotation manually during training.
Do you do some preprocess steps to make the rotations uniform?
The text was updated successfully, but these errors were encountered: