Improved GANs
Vanilla GANs prove the potential of adversarial networks. The ease of setting up models and the quality of output has sparked much interest in this field. This led to a lot of research in improving the GAN paradigm. In this section, we will cover a few of the major improvements in developing GANs.
Deep convolutional GANs
Published in 2016, the work by Radford et al. on deep convolutional GANs (DCGANs) introduced several key contributions to improve GAN outputs, apart from focusing on convolutional layers. The original GAN paper also talks about using convolutional layers, but this work discusses using deeper architectures for the same. Figure 12.10 showcases the generator architecture for a DCGAN (as proposed by the authors). The generator takes the noise vector as input and then passes it through a repeating setup of upsampling layers, convolutional layers, and batch normalization to stabilize the training.
Figure 12.10: DCGAN generator architecture...
