Potential Flow Generator with $L_2$ Optimal Transport Regularity for Generative Models
This work addresses the challenge of enhancing generative models for image translation with unpaired data, though it appears incremental as it builds on existing GANs and flow-based models.
The authors tackled the problem of improving generative models by proposing a potential flow generator with L2 optimal transport regularity, which they demonstrated to be effective in 2D problems and image translation tasks on MNIST and CelebA datasets.
We propose a potential flow generator with $L_2$ optimal transport regularity, which can be easily integrated into a wide range of generative models including different versions of GANs and flow-based models. We show the correctness and robustness of the potential flow generator in several 2D problems, and illustrate the concept of "proximity" due to the $L_2$ optimal transport regularity. Subsequently, we demonstrate the effectiveness of the potential flow generator in image translation tasks with unpaired training data from the MNIST dataset and the CelebA dataset.