Reproduce “PuzLM: Solving Jigsaw Puzzles with Sequence-to-Sequence Language Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Gur Elkin, Ofir Itzhak Shahar, Ohad Ben-Shahar
PuzLM converts each square puzzle piece into discrete tokens derived from image patches, retains border tokens, and uses a standard encoder–decoder language model to predict piece positions from the resulting sequence. On JPwLEG-5, it reports 72.1% absolute accuracy and 32.5% perfect accuracy, versus VLHSA’s 66.9% and 19.0%. On JPwLEG-3, FCViT outperforms PuzLM, so the result does not establish universal superiority.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.