← back to paper
arxiv: 2608.02109 · 2 revisions
Same Semantics, Different Paths: Self-Improving Alignment for Vision-Text Compression