DIP: Unsupervised Dense In-Context Post-training of Visual Representations Paper โข 2506.18463 โข Published 3 days ago โข 17