Textmeddelande: Looking beyond single images for contrastive semantic segmentation learning