[2002.04724] Improved Consistency Regularization for GANscontact arXivarXiv Twitter

Recent work has increased the performance of Generative Adversarial Networks (GANs) by enforcing a consistency cost on the discriminator. We improve on this technique in several ways. We first show that consistency regularization can introduce artifacts into the GAN samples and explain how to fix this issue. We then propose several modifications to the consistency regularization procedure designed to improve its performance. We carry out extensive experiments quantifying the benefit of our improvements. For unconditional image synthesis on CIFAR-10 and CelebA, our modifications yield the best known FID scores on various GAN architectures. For conditional image synthesis on CIFAR-10, we improve the state-of-the-art FID score from 11.48 to 9.21. Finally, on ImageNet-2012, we apply our technique to the original BigGAN model and improve the FID from 6.66 to 5.38, which is the best score at that model size.

7 mentions: @ak92501@gstsdn@AlexGDimakis@AkiraTOSEI@ZhengliZhao@AkiraTOSEI
Keywords: gan
Date: 2020/02/13 20:21

Referring Tweets

@AkiraTOSEI t.co/YHe74Rlc1C GANにおけるconsistency regularization(CR)の掛け方を改善。先行研究では、DのみでCRを実施していたが、D/G両方でCRを実施し、画像だけでなく潜在空間でもCRをとるImproved Consistensy Regularizationを提案。単純なアルゴリズムだが、FIDが有意に改善する。 t.co/jYjoVOmMNl
@ZhengliZhao Our new work on improving consistency-regularization techniques for GANs. Kudos to my awesome mentors and collaborators @GoogleAI. t.co/mONSv9nvsL t.co/uGd9wiSbmD
@AkiraTOSEI t.co/YHe74Rlc1C Improvement of consistency regularization (CR) in GAN. In previous research, CR was performed only for D, but in this work CR is performed not only in both D and G, but also in latent space. Simple algorithm, but significantly improves FID. t.co/7MpA9MNGM2
@ak92501 Improved Consistency Regularization for GANs pdf: t.co/IJki7WDD9T abs: t.co/h1BPZ0KKWd t.co/mDLyq1NbYS

Related Entries

Read more [2001.07966] ImageBERT: Cross-modal Pre-training with Large-scale Weak-supervised Image-Text Datacon...
0 users, 8 mentions 2020/01/30 00:52
Read more [2002.06224] Top-K Training of GANs: Improving Generators by Making Critics Less Criticalcontact arX...
0 users, 5 mentions 2020/02/23 03:50
Read more [2003.03497] MatchingGAN: Matching-based Few-shot Image Generationcontact arXivarXiv Twitter
0 users, 3 mentions 2020/03/10 14:21
Read more [2007.03898] NVAE: A Deep Hierarchical Variational Autoencoderopen searchopen navigation menucontact...
0 users, 14 mentions 2020/07/10 00:53
Read more [2008.00951] Encoding in Style: a StyleGAN Encoder for Image-to-Image Translationopen searchopen nav...
0 users, 3 mentions 2020/08/11 14:21