Exactly, the convnet is applied on 2d pixels instead of on 3D point clouds. If one have labelled 3D points (e.g. if a point is part of a car or not), one can train a network for recognition directly on 3D points.
Nice. I wonder if this could be used for robot navigation. The segmented point clouds especially are useful.
Been looking forward to a DIY Roomba project using point clouds and neural nets for navigation
So the segmentation convnet is applied on the 2d image pixels before SfM, rather than on the point cloud data? I've been looking for neural net semantic segmentation for 3D point cloud data