Reproduce “Spatially Adaptive Noise Injection” 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
Frantzeska Lavda, Maciej Falkiewicz, Van Khoa Nguyen, Alexandros Kalousis
SANI is an inference-time sampling method for diffusion image generation that estimates pixel-wise uncertainty and uses it to gate exploratory noise across an image. It preserves variance compensation while coupling noise and predicted-direction contributions. On CelebA-64 at 10 steps, SANI-G reports FID 17.09 versus 20.54 for DDIM and 36.69 for DDPM, but OCM scores 10.28.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.