Vision-based autonomous driving requires classifying each pixel as corresponding to road or not, which can be addressed using semantic segmentation. Semantic segmentation works well when used with a fully supervised model, but in practice, the required work of creating pixel-wise annotations is very expensive. Although →