Unified Mask Embedding and Correspondence Learning for Self-Supervised Video Segmentation | Read Paper on Bytez