Check output dimension

[ShaoxiongYao]

Hi,

Thank you for the amazing codebase!

I have a question regarding the implementation. I noticed that the model currently outputs 27 dimensions, which corresponds to 3 xyz translation dimensions and 24 joints of the 5-finger Shadow Hand. However, based on my understanding of the paper, the diffusion should occur in a 33-dimensional space: 3 for translation, 6 for rotation representation, and 24 for joint angles. Could you clarify whether I may be misunderstanding something?

Thank you for your help in advance.

[ymzhong66]

The rotation dimensions are handled differently in this implementation. I transferred the rotation to the object frame. I rotate the object arbitrarily and then apply the inverse transformation to the Shadow Hand. This is why the output is 27 dimensions instead of 33.