Looking Beyond Single Images for Contrastive Semantic Segmentation Learning Supplementary Material