Frederick remarked "It was super nice with your generator to be able to generate training data! It proved very useful."
The thesis explored how deep learning on 3D structures instead of still images can be achieved.
The Shape Generator was used to create "as many models as needed" and "could generate all kinds of shapes." The level of detail could be controlled so that the models were neither too simple or too complex.
In this instance, the models created were non-symmetrical "animal-like shapes" and are available on Git Hub.