Reproduce “PixARG: Pixel-Asynchronous Retrieval-Augmented Diffusion for Fast and Realistic Image Generation” 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
Zewei Xin, Qinya Li, Chaoyue Niu, Fan Wu
PixARG generates prompt-based images by retrieving an example, mapping patch-level text–image relevance, and preserving aligned regions in a spatially varying-noise starting state while regenerating less relevant regions. On LAION with SD 1.4 and DDIM, it uses 15 rather than 50 denoising steps and takes 0.367 rather than 0.883 seconds per image end to end. CLIP alignment, precision, and FID improve, while recall falls.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.