-
A Style-Based Generator Architecture for Generative Adversarial Networks
- propose an alternative generator architecture from the idea of style transfer
- Style transfer: modifying the style of an image while still preserving its content (like a dog photo in a cubic style)
- Difference from style-transfer: the spatially invariant style y is calculated from a vector from an intermediate latent space instead of an image
- A mapping network is used to map the intermediate latent space and the input latent space.
-
GAN Dissection: Visualizing and Understanding Generative Adversarial Networks *
-
[Difference-Seeking Generative Adversarial Network - Unseen Sample Geneartion] *