Tensorflow implementation of a Deep Convolutional Adversarial Network. It is used to resize Cifar-10 images from a 16x16 shape to the original 32x32. This code is designed to run on any laptop with a tensorflow-gpu installation.
Model
The Gan consists of a Deep Convolutional Network trained at distinguishing Cifar original 32x32 images from the generated ones (the Discriminator), and a Deep Deconvolutional Network tained fooling the discriminator: the Generator.
Discriminator Model
Generator Model
Results
We are aware that our model is too small to achieve any significant results. Truth is we lack the computing power for a larger one.
Nonetheless, we trained it on a 940mx Nvidia GPU.
Here are some examples of generated images:
A Deer:
A Car:
