In this task, you should train a conditional GAN to generate the right half of MNIST images, given the left half.

Load the dataset¶

Load the MNIST dataset bellow. You can use either torchvision.datasets.MNIST or sklearn.datasets.fetch_openml() or any other way to load the dataset.

Design your model¶

Implement your GAN below using torch.nn modules. Feel free to add extra cells. The generator should construct right half of an image given the left half of it and a noise. The general structure of this GAN is shown below.

Train your GAN¶

Write the training process below. Feel free to add extra cells.

Plot Generator/Discriminator losses¶

Draw examples¶

Draw real vs. generated images below (from the best epoch).