| title | JigShape: Evaluating Visual-Geometric Reasoning in VLMs through Jigsaw Puzzles |
|---|---|
| authors | Shawn Li, Wei Yang, Jike Zhong, Jiate Li, Jiawei Yang, You Qin, Ryan Rossi, Franck Dernoncourt, Roger Zimmermann, Yue Wang, Zhengzhong Tu, Vicente Ordonez, Mohit Bansal, Yue Zhao |
| date | 2026-08-04 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2607.27670 |
| upvotes | 4 (Hugging Face Daily Papers, 2026-11-01) |
| links | abs pdf hf |
| abstract | Jigsaw puzzle solving requires jointly reasoning about visual content and geometric constraints, yet existing benchmarks use rectangular cuts that create ambiguous ground truth in texture-repeated regions. We introduce \ours{}, a benchmark with tab-and-blank interlocking pieces where geometric constraints provide strong local compatibility requirements that, combined with visual content, yield unambiguous ground truth. Across 95K instances at four grid densities (4times4 to 16times16), we find that zero-shot VLMs largely lack geometric reasoning: only one of five frontier models (GPT-5.5) exceeds random baseline on 4times4 puzzles, while all others perform at chance level. While supervised fine-tuning achieves >97\% on 4times4, all models collapse on larger grids: GPT-5.5 drops from 70\% to near-random on 8times8, and even fine-tuned models fall below 5\% on 12times12. This ``scaling cliff'' suggests current architectures cannot maintain consistent constraint satisfaction as the number of pieces increases. establishes scalable geometric reasoning as an open challenge for vision-language models. |
cite
@misc{li2026jigshape,
title={JigShape: Evaluating Visual-Geometric Reasoning in VLMs through Jigsaw Puzzles},
author={Shawn Li and Wei Yang and Jike Zhong and Jiate Li and Jiawei Yang and You Qin and Ryan Rossi and Franck Dernoncourt and Roger Zimmermann and Yue Wang and Zhengzhong Tu and Vicente Ordonez and Mohit Bansal and Yue Zhao},
year={2026},
eprint={2607.27670},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2607.27670}
}