DeepAI AI Chat
Log In Sign Up

Stabilization of generative adversarial networks via noisy scale-space

by   Kensuke Nakamura, et al.
Chung-Ang University

Generative adversarial networks (GAN) is a framework for generating fake data based on given reals but is unstable in the optimization. In order to stabilize GANs, the noise enlarges the overlap of the real and fake distributions at the cost of significant variance. The data smoothing may reduce the dimensionality of data but suppresses the capability of GANs to learn high-frequency information. Based on these observations, we propose a data representation for GANs, called noisy scale-space, that recursively applies the smoothing with noise to data in order to preserve the data variance while replacing high-frequency information by random data, leading to a coarse-to-fine training of GANs. We also present a synthetic data-set using the Hadamard bases that enables us to visualize the true distribution of data. We experiment with a DCGAN with the noise scale-space (NSS-GAN) using major data-sets in which NSS-GAN overtook state-of-the-arts in most cases independent of the image content.


page 8

page 11

page 17

page 18


Are High-Frequency Components Beneficial for Training of Generative Adversarial Networks

Advancements in Generative Adversarial Networks (GANs) have the ability ...

Intriguing properties of synthetic images: from generative adversarial networks to diffusion models

Detecting fake images is becoming a major goal of computer vision. This ...

Generative Adversarial Networks for Multimodal Representation Learning in Video Hyperlinking

Continuous multimodal representations suitable for multimodal informatio...

Catch Me If You GAN: Using Artificial Intelligence for Fake Log Generation

With artificial intelligence (AI) becoming relevant in various parts of ...

Stabilizing GANs with Octave Convolutions

In this preliminary report, we present a simple but very effective techn...

Explicit Use of Fourier Spectrum in Generative Adversarial Networks

Generative Adversarial Networks have got the researchers' attention due ...

Evaluating Generative Adversarial Networks on Explicitly Parameterized Distributions

The true distribution parameterizations of commonly used image datasets ...