Some questions about the implementation.

[hehulingxiao]

Thanks for your great work，may I ask you some questions？

1. Is self.model_pose (context encoder) used to encode the context features as condition for diffuser or used to initialize the xyz coordinates during inference?
2. Are uv coordinates used to instruct the forward diffusion process for ground truth xyz coordinates or used as condition for diffuser?
3. How is the implementation of z initialization? Is it assumed to be distributed like N(0, I) in the code as no kernel mean and variance are provided like uv?

Looking forward to your reply.