OctNet

@inproceedings{riegler2017octnet,
  title={Octnet: Learning deep 3d representations at high resolutions},
  author={Riegler, Gernot and Osman Ulusoy, Ali and Geiger, Andreas},
  booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
  pages={3577--3586},
  year={2017}
}

Use an Octree as 3d representation
- Max depth = 3
Have a Neural Network work with them
Octree will use much less space as a dense voxel grid

Convolutions on Octrees: Neighbor lookups not trivial

No generational tasks possible, because it changes the structure of the octree (?)