Vision-based perception for autonomous driving requires an explicit modeling of a 3D space
where 2D latent representations are mapped and subsequent 3D operators are applied …