Comment by konstantina_ps

Hi 4thstreet, thanks for your question! Yes, we train on 2D-3D pairs. For example, we have 3D models and then we take images of them so we learn how objects look from different points of view :)