← back to paper
arxiv: 2605.10893 · 2 revisions
Grounded or Guessing? LVLM Confidence Estimation via Blind-Image Contrastive Ranking