[2009.05475] Adversarial score matching and improved sampling for image generationopen searchopen navigation menucontact arXivsubscribe to arXiv mailings

Denoising score matching with Annealed Langevin Sampling (DSM-ALS) is a recent approach to generative modeling. Despite the convincing visual quality of samples, this method appears to perform worse than Generative Adversarial Networks (GANs) under the Fréchet Inception Distance, a popular metric for generative models. We show that this apparent gap vanishes when denoising the final Langevin samples using the score network. In addition, we propose two improvements to DSM-ALS: 1) Consistent Annealed Sampling as a more stable alternative to Annealed Langevin Sampling, and 2) a hybrid training formulation,composed of both denoising score matching and adversarial objectives. By combining both of these techniques and exploring different network architectures, we elevate score matching methods and obtain results competitive with state-of-the-art image generation on CIFAR-10.

5 mentions: @jm_alexia@ak92501@Montreal_AI@hillbig@hillbig
Date: 2020/09/14 23:22

Referring Tweets

@hillbig DSM-ALSは観測に加えたノイズを予測することで対数尤度勾配を推定し、それを使ってサンプリングする。従来手法のノイズスケールを修正し、サンプル計算時に経験ベイズ平均を使ってデノイジングし、[Ho et al. 2020]で使われているネットワークを使うと品質を大きく改善できる t.co/PMiILenrHp
@hillbig Denoising Score Matching with Annealed Langevin Sampling [Song 2019, 2020] can be improved significantly by 1) correcting noise scale 2) Denoising using empirical Bayes mean at final sampling 3) Adversarial loss 4) Network Architecture used in [Ho 2020] t.co/PMiILeF2yX
@jm_alexia New paper on adversarial😠 score matching and an alternative to Langevin Sampling for better generative models! 😸 We show how we can obtain results better than SOTA GANs. 😻 Blog: t.co/Lqje2m1jvr Paper: t.co/dkxgDRKcCP Code: t.co/ajc3TkoomS
@Montreal_AI Adversarial score matching and improved sampling for image generation Jolicoeur-Martineau et al.: t.co/ZQFUIclcWe Blog : t.co/yL5LjwYJqt Code : t.co/dAF7nHgR0q #AnnealedLangevinSampling #GenerativeAdversarialNetworks #GenerativeModeling t.co/ETWwR7XG1e

Related Entries

Read more [2003.10580] Meta Pseudo Labelscontact arXivarXiv Twitter
0 users, 6 mentions 2020/03/25 21:50
Read more [2004.13324] Learning Feature Descriptors using Camera Pose Supervisionopen searchopen navigation me...
0 users, 5 mentions 2020/04/30 00:51
Read more [2007.06965] Automated Synthetic-to-Real Generalizationopen searchopen navigation menucontact arXivs...
0 users, 5 mentions 2020/07/17 00:52
Read more Free View Synthesis - YouTube
0 users, 7 mentions 2020/08/11 20:22
Read more Meta-Sim2
0 users, 5 mentions 2020/08/25 23:21