← back to paper
arxiv: 2606.06714 · 2 revisions
Anchored, Not Graded: Vision-Language Models Fail at Slant-from-Texture Perception