Skip to yearly menu bar Skip to main content


Poster
in
Workshop: Structured Probabilistic Inference and Generative Modeling

Regularized Distribution Matching Distillation for One-step Unpaired Image-to-Image Translation

Denis Rakitin · Ivan Shchekotov · Dmitry Vetrov

Keywords: [ Optimal Transport ] [ unpaired image-to-image translation ] [ diffusion distillation ] [ distribution matching ] [ variational score distillation ]


Abstract:

Diffusion distillation methods aim to compress the diffusion models into efficient one-step generators while trying to preserve quality. Among them, Distribution Matching Distillation (DMD) offers a suitable framework for training general-form one-step generators, applicable beyond unconditional generation. In this work, we introduce its modification, called Regularized Distribution Matching Distillation, applicable to unpaired image-to-image problems. We demonstrate its empirical performance in application to several translation tasks, including 2D examples and I2I between different image datasets, where it performs on par or better than multi-step diffusion baselines.

Chat is not available.