16.03.2021 Views

Advanced Deep Learning with Keras

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Improved GANs

Since the introduction of the Generative Adversarial Networks (GANs) in 2014[1],

its popularity has rapidly increased. GANs have proved to be a useful generative

model that can synthesize new data that look real. Many of the research papers

in deep learning that followed, proposed measures to address the difficulties

and limitations of the original GAN.

As we discussed in previous chapters, GANs can be notoriously difficult to train

and prone to mode collapse. Mode collapse is a situation where the generator is

producing outputs that look the same even though the loss functions are already

optimized. In the context of MNIST digits, with mode collapse, the generator

may only be producing digits 4 and 9 since they look similar. Wasserstein GAN

(WGAN)[2] addressed these problems by arguing that stable training and mode

collapse can be avoided by simply replacing the GAN loss function based on

Wasserstein 1 or Earth-Mover distance (EMD).

However, the issue of stability is not the only problem of GANs. There is also

the increasing need to improve the perceptive quality of the generated images.

Least Squares GAN (LSGAN)[3] proposed to address both these problems

simultaneously. The basic premise is that sigmoid cross entropy loss leads to

a vanishing gradient during training. This results in poor image quality. Least

squares loss does not induce vanishing gradients. The resulting generated images

are of higher perceptive quality when compared to vanilla GAN generated images.

In the previous chapter, CGAN introduced a method for conditioning the output

of the generator. For example, if we wanted to get digit 8, we would include the

conditioning label in the input to the generator. Inspired by CGAN, the Auxiliary

Classifier GAN (ACGAN)[4] proposed a modified conditional algorithm that

results in better perceptive quality and diversity of the outputs.

[ 125 ]

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!