2017-03-31 · preprint · Generative Adversarial Networks and Image Synthesis · en
highly citedfrom the source record
Generative Adversarial Networks (GANs) are powerful generative models, but suffer from training instability. The recently proposed Wasserstein GAN (WGAN) makes progress toward stable training of GANs, but sometimes can still generate only low-quality samples or fail to converge. We find that these problems are often due to the use of weight clipping in WGAN to enforce a Lipschitz constraint on the critic, which can lead to undesired behavior. We propose an alternative to clipping weights: penalize the norm of gradient of the critic with respect to its input. Our proposed method performs better than standard WGAN and enables stable training of a wide variety of GAN architectures with almost no hyperparameter tuning, including 101-layer ResNets and language models over discrete data. We also achieve high quality generations on CIFAR-10 and LSUN bedrooms.
Claims drawn from cited facts, not live model generation.
This paper concerns generative adversarial networks and image synthesis, and it is a preprint. Because it is a preprint, it has not yet cleared peer review, so its findings should be weighted with corresponding caution.
3 cited facts
This work is the product of a small team of 3 researchers.
1 cited fact
The work has received 6,799 citations, marking it as already an established reference point in its field. It draws on 33 referenced works, giving the citation footprint a clear sense of its scholarly reach.
3 cited facts
Improved Training of Wasserstein GANs is a preprint, focusing on Generative Adversarial Networks and Image Synthesis, published in 2017.
Improved Training of Wasserstein GANs has 6,799 citations in our record, placing it among the highly cited works we track.
Our record lists 3 authors for Improved Training of Wasserstein GANs and references 33 other works. Author names aren't in our data yet — the source link on the page lists the full byline.
Source facts, citations, and refresh stamp for this record.
Sources: openalex_works