gengshan-y
commented
2 years ago Hi, thanks for the questions.
- near-far (N,2): near and far plane for each image.
- obj_bound (3): a loose bound of the object surface along xyz axis in the canonical space. see B.3 of the paper for details.
- obj_scale: This is a fixed constant and should always equals to 10. Please ignore it.
- bound_factor: The looseness of the initial near_far plane. To cover more spatial points than the initial unit sphere during rendering, we set the near-far planes with a bigger sphere as
bound_factor* initial unit sphere. We found bound_factor=2 to be a good value and fixed it through experiments.
The reason near_far are nn.parameters is to facilitate model loading (there may be a better design). Indeed, near_far are not optimized through gradients, but only updated using rest shape and cameras here.
Hi, thanks again for your awesome work!
Could you please tell me the means and relationships of parameter 'near_far', 'obj_bound', 'obj_scale', 'bound', and 'bound_factor'? I am so confused about these parameters.
Also, I can understand you set the object center localized at z=0.3 in the world space, and design the 'warmp_shape' to initialize the object as a small sphere by training the SDF. But why set the near_far (initialized by 0-0.6) as a learning parameter (reset_nf) instead of a fixed hyperparameter?