Reproduce “CounterFlow: Target-Conditioned Flow Matching for Visual Counterfactuals” 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
Paul Sevellec, Elisa Fromont, Romaric Gaudel, Laurence Roze, Paul Viallard, Laetitia Houeix
CounterFlow generates image counterfactuals by learning a transport conditioned on source and target classes, rather than repeatedly steering a generative model. It trains on paired images and integrates the learned field at inference using 40 Euler steps. On CIFAR-10, it reaches 100.0% conditional validity and improves FID and compute measures over C3LT; evidence is limited to low-resolution MNIST and CIFAR-10.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.