Notebook: DCGAN, SVHN

Now you have all the information you need to implement a deep convolutional GAN, capable of generating more complex images! The next notebook is all about building a DCGAN that can generate new images that look like house addresses from the Google Streetview dataset, SVHN.

It's suggested that you open the notebook in a new, working tab and continue working on it as you go through the instructional videos in this tab. This way you can toggle between learning new skills and coding/applying new skills.

To open this notebook, you have two options:

• Go to the next page in the classroom (recommended).
• Clone the repo from Github and open the notebook DCGAN_Exercise.ipynb in the dcgan-svhn folder. You can either download the repository with git clone https://github.com/udacity/deep-learning-v2-pytorch.git, or download it as an archive file from this link.

Instructions

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

This is a self-assessed lab. If you need any help or want to check your answers, feel free to check out the solutions notebook in the same folder, or by clicking here.

GPU Workspaces

The next workspace is GPU-enabled, which means you can select to train on a GPU instance. The recommendation is this:

• Load in data, test functions and models (checking parameters and doing a short training loop) while in CPU (non-enabled) mode
• When you're ready to extensively train and test your model, enable GPU to quickly train the model!

All models and data they see as input will have to be moved to the GPU device, so take note of the relevant movement code in the model creation and training process.