Vision unified representation model with standard and compressed features for classification, generation, and more: https://arxiv.org/abs/2601.14256