Reproduce “Support-guided OOD Removal with Diffusion” 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
Alireza Kheirandish, Sara Fridovich-Keil
SORD reconstructs localized structured corruption using a support mask and a diffusion prior for the desired clean-signal class. It initializes candidates by inpainting, then iteratively reuses corrupted measurements while refining residuals. On MNIST digits overlaid on CelebA, its corrupted-region pixel-reconstruction score is 27.16 ± 2.04 dB, versus 22.51 ± 3.89 for masked SGM; results depend on an accurate mask.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.