Reproducing the toy example in "Calibrated Adversarial Refinement for Multimodal Semantic Segmentation" by Kassapis et al.