StarGAN

Definition

StarGAN is a GAN model designed for multi-domain image-to-image translation tasks. Unlike previous models that required separate networks for each domain pair, StarGAN can perform image translations across multiple domains using a single generator network.

Architecture

The generator takes an input image and a target domain label to produce the translated image, and the discriminator network that not only distinguishes between real and fake images but also classifies the domain of the input image.

Mask Vector

StarGAN introduces a mask vector to handle datasets with partial domain labels, allowing it to ignore unspecified labels during training.

Objective Function

The objective function of StarGAN consists of three losses: adversarial loss, domain classification loss, and reconstruction loss. The adversarial loss ensures the generated images are realistic, the domain classification loss helps the model learn domain-specific features, and the reconstruction loss ensures the original image can be reconstructed when translating back to the source domain.

Adversarial loss: $L_{\text{adv}}(G,D)=E_{x\sim p_r(x)}[\log D_{\text{src}}(x)]+E_{x\sim p_r(x)}[\log (1-D_{\text{src}}(G(x,c)))]$ Domain classification loss: $L_{\text{cls}}(G,D)=E_{x\sim p_r(x)}[-\log D_{\text{cls}}(c^{\prime }|x)]+E_{x\sim p_r(x)}[-\log D_{\text{cls}}(c|G(x,c))]$ Reconstruction loss: $L_{\text{rec}}(G)=E_{x\sim p_r(x)}[||x-G(G(x,c),c^{\prime })|{|}_1]$ where:

• $G$ is the generator
• $D$ is the discriminator
• $D_{\text{src}}$ is the discriminator’s source prediction
• $D_{\text{cls}}$ is the discriminator’s domain classification
• $p_r$ is the distribution of the input data
• $c$ is the target domain label.
• $c^{\prime }$ is the original domain label.

The full objective function for StarGAN can be written as: ${\min }_{{\theta }_G}{\max }_{{\theta }_D}L=L_{\text{adv}}(G,D)+{\lambda }_{\text{cls}}{L}_{\text{cls}}(G,D)+{\lambda }_{\text{rec}}{L}_{\text{rec}}(G)$ where:

• ${\theta }_G$ and ${\theta }_D$ are the parameters of the generator and discriminator, respectively
• ${\lambda }_{\text{cls}}$ and ${\lambda }_{\text{rec}}$ are hyperparameters controlling the importance of each loss term