Implementation for progressive training of convex Wasserstein GANs, from our paper, Hidden Convexity of Wasserstein GANs: Interpretable Generative Models with Closed-Form Solutions. Built on theory that demonstrates that GANs with linear generators and two-layer quadratic-activation neural network discriminators are convex programs and can be solved in closed-form, and motivated by the success of progressive training of GANs, we show that the convex formulation improves upon baseline heuristic stochastic Gradient Descent-Ascent (GDA), which is typically used in practice. This repository includes implementations of the convex form and the baseline on the CelebA faces dataset.
-
Notifications
You must be signed in to change notification settings - Fork 1
ardasahiner/ProCoGAN
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Folders and files
Name | Name | Last commit message | Last commit date | |
---|---|---|---|---|
Repository files navigation
About
Progressive Training of Convex GANs.
Resources
Stars
Watchers
Forks
Releases
No releases published