← back to paper
arxiv: 2605.12325 · 2 revisions
VIP: Visual-guided Prompt Evolution for Efficient Dense Vision-Language Inference