[neuronexmachina]

Isn't the typical training data used in self-driving basically things like object labeling/segmentation and motion prediction? I'm not sure why that would be significantly different for visual vs depth-map data.