← back to paper
arxiv: 2607.27670 · 2 revisions
JigShape: Evaluating Visual-Geometric Reasoning in VLMs through Jigsaw Puzzles