Question about grad_feat, grad_color and grad_dir.

[ghost]

Hi, Thanks for sharing this work.

As neural points features are extracted from MVSNet with 2D feature extractor and they are pretrained.
Then why do we need to set feat_grad to true? It may drastically increase memory capacity.
(Also I don't know why we need to set color_grad and dir_grad to True as well.)

Thanks again.