In #115 the ability to chunk up the scene for aggregation was removed. So was using PyTorch3D to compute pixel to face correspondences. Both of these should likely be implemented in their own derived mesh class. It is important to make the PyTorch3D dependency optional, since it is a challenging package to install and avoiding this headache was one of the main reasons we implemented an alternative approach.
In #115 the ability to chunk up the scene for aggregation was removed. So was using
PyTorch3D
to compute pixel to face correspondences. Both of these should likely be implemented in their own derived mesh class. It is important to make the PyTorch3D dependency optional, since it is a challenging package to install and avoiding this headache was one of the main reasons we implemented an alternative approach.