Reproduce “Benchmarking Content-Based Puzzle Solvers on Corrupted Jigsaw Puzzles” 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

Richard Dirauf, Florian Wolz, Dario Zanca, Björn Eskofier

This study benchmarks State-Of-The-Art content-based jigsaw puzzle solvers against various visual corruptions mimicking real-world artifact degradation. It reveals that while deep learning models initially struggle with corrupted data, fine-tuning them with augmented data substantially improves their robustness, with fine-tuned Positional Diffusion consistently emerging as the most effective method in most corrupted scenarios.

Key insights

  • Heuristic solvers exhibited rapid performance degradation with increasing corruption, especially when pixel values were affected, highlighting their sensitivity to direct content alterations.
  • Deep learning models, while initially not robust to corruptions, demonstrated significant adaptability; fine-tuning them with augmented data consistently and substantially improved their resilience across all corruption types.
  • The fine-tuned Positional Diffusion model emerged as the most robust solver for corrupted scenarios, suggesting that Diffusion Probabilistic Models are a promising architecture for handling fragmented data.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.