It predicts all the same orientation angle

Thanks for the wonderful work.

I've tried training the model and it seemed successfully worked.
But when I ran the 3D orientation regression part, it appeared that always same angle is predicted.
I found a weight that you provided from one of the issues, and it seems it correctly predicts the angle when I used the weight.
Thus, it is considered the training has a problem.

Have you ever had this kind of issue? Thank you in advance.