16.03.2021 Views

Advanced Deep Learning with Keras

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Improved GANs

Unlike CGAN, the sample outputs appearance does not vary widely during training.

The MNIST digit image perceptive quality is also better. Figure 5.3.4 shows a side by

side comparison of every MNIST digit produced by both CGAN and ACGAN. Digits

2-6 are of better quality in ACGAN than in CGAN.

I encourage you to run the trained generator model to see new synthesized MNIST

digits images:

python3 acgan-mnist-5.3.1.py --generator=acgan_mnist.h5

Alternatively, a specific digit (for example, 3) to be generated can also be requested:

python3 acgan-mnist-5.3.1.py --generator=acgan_mnist.h5 --digit=3

Figure 5.3.3: The sample outputs generated by the ACGAN as a function of train steps

for labels [0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5]

[ 158 ]

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

Saved successfully!

Ooh no, something went wrong!