dcgan-svhn

Deep Convolutional GANs for the Google Streetview SVHN dataset. The code is based on the miniproject of the chapter on GAN of the Udacity Deep Learning course.

Content

The notebook implements the following steps

• Load in SVHN data
• Pre-process the data scaling the pixel to a desired range
• Define two adversarial networks, a discriminator and a generator, that utilize convolutional or transpose convolutional layers
• Train the networks and generate new images

Features

If available, PyTorch trains on GPU.

References

Udacity github