Reproduce “CODE: Confident Ordinary Differential Editing” 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
Bastien van Delft, Tommaso Martorella, Alexandre Alahi
CODE edits damaged images without being told their corruption type, using a pretrained diffusion model to adjust images while preserving input fidelity. It encodes the image with a probability-flow ODE, applies stochastic Langevin updates, and reverses the encoding. On synthetic CelebA-HQ corruptions, CODE reports FID 30.66 versus SDEdit’s 47.84; scores use the best four samples selected against clean sources.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.