What is Deep Convolution Generative Adversarial Network(DCGAN)

Deep Convolution Generative Adversarial Network or DCGAN is a unsupervised representation of learning GANs. In deep convolutional adversarial pair learns a hierarchy of representations from object parts to scenes in both the generator and discriminator. In short we say, GANs with CNNs architecture

In DCGAN paper, They contributions-

1. They gives proper architectural of Convolutional GANs.
2. They trained discriminators for image classification tasks and showing competitive performance with other unsupervised algorithms.
3. They shows visualize the filters learnt by GANs.
4. And they also shows generators have interesting vector arithmetic properties allowing for easy manipulation of many semantic qualities of generated samples

Some DCGAN examples

Fig : Real Images vs. Fake Images

Fig : DCGAN every steps training evaluation in hand written digit

Basic structure behind the DCGANs

Fig : Basic Generator & Discriminator in DCGAN

Improved network architectural GANs to DCGANs

Resources

Basic GAN

Pytorch DCGAN

Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks

Papers with code DCGANs

DCGAN tensorflow

DCGAN faces pytorch