search

tijiang13 / InstantAvatar

333 stars 23 forks source link

Question about the generated near far in Snarf_deformer. #64

Closed zhao-yiqun closed 5 months ago

zhao-yiqun commented 5 months ago

Hi

thanks for your code releasing,

I have some questions about your code implementation here. near, far

I wonder why the near/far are generated from the norm of ray_o(for both training code and inference code)? I can not fully understand this, and hope you can help me about this.

(Based on my initial understanding, the near plane should be the nearest point from human body, and the far plane should be the farest plane that can fully cover the human body)

rays.o is the camera origin in smpl-coordinate, but, why do you set the torch.norm(rays.o) -1 for the near plane?

Thanks in advance

tijiang13 commented 5 months ago

Hi Yiqun,

Here we only need a rough estimation of the near and far plane. We can definitely make it the distance between nearest point from human body and camera but it might be a bit slower to calculate that.

Best, Tianjian

zhao-yiqun commented 5 months ago

Thanks for your kind reply,

Now I understand that.

Bests, Yiqun