Reproduce “Astrophotography turbulence mitigation via generative 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

Joonyeoup Kim, Yu Yuan, Xingguang Zhang, Xijun Wang, Stanley Chan

Purdue University researchers developed AstroDiff, a dual-branch generative model that significantly mitigates atmospheric turbulence in astrophotography by fusing data-driven restoration with domain-specific priors. It achieves a 76.2% improvement in BRISQUE scores on real astronomical images and outperforms existing methods in perceptual quality, particularly under severe turbulence conditions.

Key insights

  • Generative models, specifically diffusion models, can effectively leverage high-quality prior information unique to astronomical objects for improved image restoration.
  • A dual-branch architecture that distinctly learns generative priors and data-driven restoration is more effective than single-branch approaches for complex image degradation.
  • Bayesian Diffusion with SGLD provides an effective mechanism to fuse complementary information from generative priors and data-driven restoration for robust reconstruction.

Summary by alphaXiv.

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