Chaining the computer-vision encoder with the 3D-graphics rendering decoder provides an opportunity for unsupervised learning for computer vision using unlabelled datasets.