Generative AI with Python and TensorFlow 2/

...

/

Vanilla GAN

Vanilla GAN

Learn how to apply a basic understanding of general adversarial networks and build one from scratch.

We'll cover the following...

The vanilla GAN will consist of a repeating block architecture similar to the one presented in the original paper. We’ll try to replicate the task of generating MNIST digits using our network.

The overall GAN setup can be seen in the figure below. The figure outlines a generator model with a noise vector zz as input and repeating blocks that transform and scale up the vector to the required dimensions. Each block consists of a dense layer followed by LeakyReLU activation and a batch-normalization layer. We simply reshape the output from the final block to transform it into the required output image size.

Press + to interact
Vanilla GAN architecture
Vanilla GAN architecture

The discriminator, on the other hand, is a simple feedforward network. This model takes an image as input (a real image or the fake output from the generator) and classifies it as real or fake. This simple setup of two competing models helps us to train the overall GAN.

Implementation

We’ll be relying on TensorFlow 2 and using the high-level Keras API wherever possible. The first step is to define the discriminator model. In this implementation, we’ll use a very basic multilayer perceptron (MLP) as the discriminator model:

Press + to interact
def build_discriminator(input_shape=(28, 28,), verbose=True):
    
    """
    Utility method to build a MLP discriminator
    Parameters:
        input_shape:
            type:tuple. Shape of input image for classification.
                        Default shape is (28,28)->MNIST
        verbose:
            type:boolean. Print model summary if set to true.
                        Default is True
    Returns:
        tensorflow.keras.model object
    """
    
    model = Sequential()
    model.add(Input(shape=input_shape))
    model.add(Flatten())   
    model.add(Dense(512))
    model.add(LeakyReLU(alpha=0.2))
    model.add(Dense(256))
    model.add(LeakyReLU(alpha=0.2))
    model.add(Dense(1, activation='sigmoid'))
    
    if verbose:
        model.summary()
    return model

  • Line 17: We add an Input layer to the model with the specified input shape.

  • Line 18: We add a Flatten layer to the model. It flattens the input, which is necessary when transitioning ...