MearcAll papers

arXiv 1704.00028 · cs.LG stat.ML

Improved Training of Wasserstein GANs

Ishaan Gulrajani, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, Aaron Courville

Claims were extracted automatically by reducer-1 on 2026-11-12. The paper's license does not allow republishing it, so long claims are shortened to a quotation. Read the full text on arXiv.

No reproduction recordedclm_7fceef62
We propose an alternative to clipping weights: penalize the norm of gradient of the critic with respect to its input.

Abstract · page 1 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_795dd05e
We also achieve high quality generations on CIFAR-10 and LSUN bedrooms.†

Abstract · page 1 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_d3bff07b
On toy datasets, we demonstrate how critic weight clipping can lead to undesired behavior.

Introduction · page 1 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_5a706bf0
We propose gradient penalty (WGAN-GP), which does not suffer from the same problems.

Introduction · page 1 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_26b40427
We demonstrate stable training of varied GAN architectures, performance improvements over weight clipping, high-quality image generation, and a character-level GAN language model without any discrete sampling. ∗Now at Google Brain … (excerpt)

Introduction · page 1 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_1836b8b2
We prove these in the Appendix.

Properties of the optimal WGAN critic · page 2 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_517e2d5b
In addition to the baseline DC- GAN architecture from [22], we choose six architectures whose successful training we demonstrate: (1) no BN and a constant number of filters in the … (excerpt)

Training varied architectures on LSUN bedrooms · page 6 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_257dd77d
WGAN-GP significantly outperforms weight clipping and performs comparably to DCGAN. we train one model with the same optimizer (RMSProp) and learning rate as WGAN with weight clipping, and another model … (excerpt)

Improved performance over weight clipping · page 7 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_9a540671
When we add label information (using the method in [20]), the same architecture outperforms all other published models except for SGAN.

Sample quality on CIFAR-10 and LSUN bedrooms · page 7 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_697a36e8
Our unsupervised model achieves state-of-the-art performance, and our conditional model outperforms all others except SGAN.

Sample quality on CIFAR-10 and LSUN bedrooms · page 7 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_b6e1dc7d
For comparison, our standard DCGAN achieved a score of 7.24.

More details for training random architectures within a set · page 13 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_22b8ec0e
The two-sided penalty achieved a higher Inception score in 100 of the trials, compared to 77 for the one-sided penalty.

Experiments with one-sided penalty · page 14 · Automated extraction, not reviewed · 0 reviewed reproductions

Reproduced one of these claims?

Record your result in the Mearc app on your own machine, then use Publish this result with this paper's id (1704.00028) and the claim id shown next to the claim. Mearc creates a file that holds only what you choose to publish. Add it to the public claim data by pull request; a maintainer other than you reviews it before it counts.