Abstract:The generative adversarial network (GAN) is first proposed in 2014, and this kind of network model is machine learning systems that can learn to measure a given distribution of data, one of the most important applications is style transfer. Style transfer is a class of vision and graphics problems where the goal is to learn the mapping between an input image and an output image. CYCLE-GAN is a classic GAN model, which has a wide range of scenarios in style transfer. Considering its unsupervised learning characteristics, the mapping is easy to be learned between an input image and an output image. However, it is difficult for CYCLE-GAN to converge and generate high-quality images. In order to solve this problem, spectral normalization is introduced into each convolutional kernel of the discriminator. Every convolutional kernel reaches Lipschitz stability constraint with adding spectral normalization and the value of the convolutional kernel is limited to [0, 1], which promotes the training process of the proposed model. Besides, we use pretrained model (VGG16) to control the loss of image content in the position of l1 regularization. To avoid overfitting, l1 regularization term and l2 regularization term are both used in the object loss function. In terms of Frechet Inception Distance (FID) score evaluation, our proposed model achieves outstanding performance and preserves more discriminative features. Experimental results show that the proposed model converges faster and achieves better FID scores than the state of the art.

An Feature Image Generation Based on Adversarial Generation Network

Creative and Diverse Artwork Generation Using Adversarial Networks

Face image generation and feature visualization using deep convolutional generative adversarial networks

SPGAN: Siamese projection Generative Adversarial Networks

Realistic Face Image Generation Based on Generative Adversarial Network

Perceptual Image Quality Using Dual Generative Adversarial Network

Generate Desired Images from Trained Generative Adversarial Networks.

Self-Modulated Feature Fusion GAN for Text-to-Image Synthesis

Conditional Image Generation Using Feature-Matching Gan

Stylized Adversarial Autoencoder for Image Generation

CAGAN: Consistent Adversarial Training Enhanced GANs.

Image Generation and Recognition Technology Based on Attention Residual GAN

Comparing Generative Adversarial Network Techniques for Image Creation and Modification

Feature-Grounded Single-Stage Text-to-Image Generation

Semantics-Enhanced Adversarial Nets for Text-to-Image Synthesis.

Model-Guided Generative Adversarial Networks for Unsupervised Fine-Grained Image Generation

Generating Cartoon Images from Face Photos with Cycle-Consistent Adversarial Networks

Adversarial Code Learning for Image Generation

Image generation step by step: animation generation-image translation

TCGAN: Three-Channel Generate Adversarial Network

SCGAN: Semi-Centralized Generative Adversarial Network for Image Generation in Distributed Scenes