Image tokenizers define the ``visual language'' of unified multimodal models, yet are commonly studied through isolated metrics or generation-/understanding-…
机构:Amazon
来源:arXiv 2609.09143 | AI4Papers 论文推荐平台