Automate GAN Configuration in PyTorch

Target audience: Advanced
Estimated reading time: 5'

Posts history

Working with the architecture and deployment of Generative Adversarial Networks (GANs) often involves complex details that can be difficult to understand and resolve. Consider the advantage of identifying neural components that are reusable and can be utilized in both the generator and discriminator parts of the network.

Table of contents

The challenge

Convolutional networks

     Neural blocks

     Inverting a convolutional neural block

     Sizing de-convolutional layers

     Assembling the de-convolutional network

References

Notes:

• This post steers clear of the intricate technicalities of generative adversarial networks and convolutional neural networks. Instead, it focuses on automating the setup process for certain neural components.
• Readers are expected to have a foundational knowledge of neural networks and familiarity with the PyTorch library.
• Environments: Python 3.9, PyTorch 1.9.1

The challenge

This article is focused on streamlining the development of Deep Convolutional Generative Adversarial Networks (DCGANs) [ref 1]. We achieve this by configuring the generator in relation to the setup of the discriminator. The main area of our study is the well-known application of using GANs to differentiate between real and fake images.

For those unfamiliar with GANs..... 

Generative Adversarial Networks (GANs) [ref 2] are a type of unsupervised learning model that identify patterns within data and utilize these patterns for data augmentation, creating new samples that closely resemble the original dataset. GANs belong to the family of generative models, which also includes variational auto-encoders and maximum likelihood estimation (MLE) models. The unique aspect of GANs is that they convert the problem into a form of supervised learning by employing two competing networks:

• The Generator model, which is trained to produce new data samples.
• The Discriminator model, which aims to differentiate between real samples (from the original dataset) and fake ones (created by the Generator).

Crafting and setting up components like the generator and discriminator in a Generative Adversarial Network (GAN), or the encoder and decoder layers in a Variational Convolutional Auto-Encoder (VAE), can often be a repetitive and laborious process.

In fact, some aspects of this process can be entirely automated. For instance, the generative network in a convolutional GAN can be designed as the inverse of the discriminator using a de-convolutional network. Similarly, the decoder in a VAE can be automatically configured based on the structure of its encoder.

Functional representation of a simple deep convolutional GAN

Neural component reusability is key to generate a de-convolutional network from a convolutional network. To this purpose we break down a neural network into computational blocks.

Convolutional networks

In its most basic form, a Generative Adversarial Network (GAN) consists of two distinct neural networks: a generator and a discriminator.

Neural blocks

Each of these networks is further subdivided into neural blocks or groups of PyTorch modules, which include elements like hidden layers, batch normalization, regularization, pooling modes, and activation functions. Take, for example, a discriminator that is structured using a convolutional neural network [ref 3] followed by a fully connected (restricted Boltzmann machine) network. The PyTorch modules corresponding to each layer are organized into what we call a neural block class.

A PyTorch modules of the convolutional neural block [ref 4] are:

• Conv2d: Convolutional layer with input, output channels, kernel, stride and padding
• Dropout: Drop-out regularization layer
• BatchNorm2d: Batch normalization module
• MaxPool2d Pooling layer
• ReLu, Sigmoid, ... Activation functions

Representation of a convolutional neural block with PyTorch modules

The constructor of the neural block is designed to initialize all its parameters and modules in the correct sequence. For simplicity, we are not including regularization elements like drop-out (which essentially involves bagging of sub-networks) in this setup.

Important note: Each step of the algorithm makes reference to comments in the code (i.e.  The first step [1] is to initialize the number of input and output channels refers to  # [1] - initialize the input and output channels).

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 3 4 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49

class ConvNeuralBlock(nn.Module): def __init__(self, in_channels: int, out_channels: int, kernel_size: int, stride: int, padding: int, batch_norm: bool, max_pooling_kernel: int, activation: nn.Module, bias: bool, is_spectral: bool = False): super(ConvNeuralBlock, self).__init__() # Assertions are omitted # [1] - initialize the input and output channels self.in_channels = in_channels self.out_channels = out_channels self.is_spectral = is_spectral modules = [] # [2] - create a 2 dimension convolution layer conv_module = nn.Conv2d( self.in_channels, self.out_channels, kernel_size=kernel_size, stride=stride, padding=padding, bias=bias) # [6] - if this is a spectral norm block if self.is_spectral: conv_module = nn.utils.spectral_norm(conv_module) modules.append(conv_module) # [3] - Batch normalization if batch_norm: modules.append(nn.BatchNorm2d(self.out_channels)) # [4] - Activation function if activation is not None: modules.append(activation) # [5] - Pooling module if max_pooling_kernel > 0: modules.append(nn.MaxPool2d(max_pooling_kernel)) self.modules = tuple(modules)

We considering the case of a generative model for images. The first step [#1] is to initialize the number of input and output channels, then create the 2-dimension convolution [#2], a batch normalization module [#3] an activation function [#4] and finally a max pooling module [#5]. The spectral norm regularization term [#6] is optional.
The convolutional neural network is assembled from convolutional and feedback forward neural blocks, in the following build method.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55

class ConvModel(NeuralModel): def __init__(self, model_id: str, # [1] - Number of input and output unites input_size: int, output_size: int, # [2] - PyTorch convolutional modules conv_model: nn.Sequential, dff_model_input_size: int = -1, # [3] - PyTorch fully connected dff_model: nn.Sequential = None): super(ConvModel, self).__init__(model_id) self.input_size = input_size self.output_size = output_size self.conv_model = conv_model self.dff_model_input_size = dff_model_input_size self.dff_model = dff_model @classmethod def build(cls, model_id: str, conv_neural_blocks: list, dff_neural_blocks: list) -> NeuralModel: # [4] - Initialize the input and output size # for the convolutional layer input_size = conv_neural_blocks[0].in_channels output_size = conv_neural_blocks[len(conv_neural_blocks) - 1].out_channels # [5] - Generate the model from the sequence # of conv. neural blocks conv_modules = [conv_module for conv_block in conv_neural_blocks for conv_module in conv_block.modules] conv_model = nn.Sequential(*conv_modules) # [6] - If a fully connected RBM is included in the model .. if dff_neural_blocks is not None and not is_vae: dff_modules = [dff_module for dff_block in dff_neural_blocks for dff_module in dff_block.modules] dff_model_input_size = dff_neural_blocks[0].output_size dff_model = nn.Sequential(*tuple(dff_modules)) else: dff_model_input_size = -1 dff_model = None return cls( model_id, conv_dimension, input_size, output_size, conv_model, dff_model_input_size, dff_model)

The standard constructor [#1] sets up the count of input/output channels, along with the PyTorch modules for the convolutional layers [#2] and the fully connected layers [#3].

The class method, build, creates the convolutional model using convolutional neural blocks and feed-forward neural blocks. It determines the dimensions of the input and output layers based on the first and last neural blocks [#4], and then produces the PyTorch convolutional modules [#5] and modules for fully-connected layers [#6] from these neural blocks.

Following this, we proceed to construct the de-convolutional neural network utilizing the convolutional blocks.

Inverting a convolutional block

To build a GAN, one must follow these steps:

1. Select and specify the PyTorch modules that will constitute each convolutional layer.
2. Assemble these chosen modules into a single convolutional neural block.
3. Construct the generator and discriminator of the GAN by integrating these neural blocks.
4. Link the generator and discriminator to create a fully functional GAN.

The aim here is to create a builder capable of producing the de-convolutional network. This network will act as the GAN's generator, formulated on the basis of the convolutional network described in the preceding section.

The process begins with the extraction of the de-convolutional block from an already established convolutional block.

Conceptual automated generation of de-convolutional block

The standard constructor for the neural block in a de-convolutional network sets up all the essential parameters required for the network, with the exception of the pooling module (which is not necessary). The code example provided demonstrates how to create a De-convolutional neural block. This process involves using convolution parameters like the number of input and output channels, kernel size, stride, padding, along with batch normalization and the activation function.

class DeConvNeuralBlock(nn.Module): def __init__(self, in_channels: int, out_channels: int, kernel_size: int, stride: int, padding: int, batch_norm: bool, activation: nn.Module, bias: bool) -> object: super(DeConvNeuralBlock, self).__init__() self.in_channels = in_channels self.out_channels = out_channels modules = [] # Two dimension de-convolution layer de_conv = nn.ConvTranspose2d( self.in_channels, self.out_channels, kernel_size=kernel_size, stride=stride, padding=padding, bias=bias) # Add the deconvolution block modules.append(de_conv) # Add the batch normalization, if defined if batch_norm: modules.append(nn.BatchNorm2d(self.out_channels)) # Add activation modules.append(activation) self.modules = modules

Be aware that the de-convolution block lacks pooling capabilities. The class method named auto_build accepts a convolutional neural block, the number of input and output channels, and an optional activation function to create a de-convolutional neural block of the DeConvNeuralBlock type. The calculation of the number of input and output channels for the resulting deconvolution layer is handled by the private method __resize.

@classmethod def auto_build(cls, conv_block: ConvNeuralBlock, in_channels: int, out_channels: int = None, activation: nn.Module = None) -> nn.Module: # Extract the parameters of the source convolutional block kernel_size, stride, padding, batch_norm, activation = \ DeConvNeuralBlock.__resize(conv_block, activation) # Override the number of input_tensor channels # for this block if defined next_block_in_channels = in_channels if in_channels is not None \ else conv_block.out_channels # Override the number of output-channels for # this block if specified next_block_out_channels = out_channels if out_channels is not None \ else conv_block.in_channels return cls( conv_block.conv_dimension, next_block_in_channels, next_block_out_channels, kernel_size, stride, padding, batch_norm, activation, False)

Sizing de-convolutional layers

The next task consists of computing the size of the component of the de-convolutional block from the original convolutional block. 

@staticmethod
def __resize(
  conv_block: ConvNeuralBlock,
  updated_activation: nn.Module) -> (int, int, int, bool, nn.Module):
  conv_modules = list(conv_block.modules)
    
  # [1] - Extract the various components of the 
  #        convolutional neural block
  _, batch_norm, activation = DeConvNeuralBlock.__de_conv_modules(conv_modules)
  

  # [2] - override the activation function for the 
  #        output layer, if necessary
  if updated_activation is not None:
     activation = updated_activation
    
    # [3]- Compute the parameters for the de-convolutional 
    #       layer, from the conv. block
     kernel_size, _ = conv_modules[0].kernel_size
     stride, _ = conv_modules[0].stride
     padding = conv_modules[0].padding

 return kernel_size, stride, padding, batch_norm, activation

The __resize method performs several functions: it retrieves the PyTorch modules for the de-convolutional layers from the initial convolutional block [#1], incorporates the activation function into the block [#2], and ultimately sets up the parameters for the de-convolutional layer [#3].

Additionally, there's a utility method named __de_conf_modules. This method is responsible for extracting the PyTorch modules associated with the convolutional layer, the batch normalization module, and the activation function for the de-convolution, all from the convolution's PyTorch modules.

@staticmethod
def __de_conv_modules(conv_modules: list) -> \
        (torch.nn.Module, torch.nn.Module, torch.nn.Module):

  activation_function = None
  deconv_layer = None
  batch_norm_module = None

  # 4- Extract the PyTorch de-convolutional modules 
  #     from the convolutional ones
  for conv_module in conv_modules:
     if DeConvNeuralBlock.__is_conv(conv_module):
         deconv_layer = conv_module
     elif DeConvNeuralBlock.__is_batch_norm(conv_module):
         batch_norm_moduled = conv_module
     elif DeConvNeuralBlock.__is_activation(conv_module):
        activation_function = conv_module

  return deconv_layer, batch_norm_module, activation_function

Convolutional layers

and the height of the two dimension output data is

De-convolutional layers

As expected, the formula to compute the size of the output of a de-convolutional layer is the mirror image of the formula for the output size of the convolutional layer.

and

Assembling de-convolutional network

Finally, a de-convolutional model, categorized as DeConvModel, is constructed using a sequence of PyTorch modules, referred to as de_conv_model. The default constructor [#1] is used once more to establish the dimensions of the input layer [#2] and the output layer [#3]. It also loads the PyTorch modules, named de_conv_modules, for all the de-convolutional layers.

class DeConvModel(NeuralModel, ConvSizeParams):

  def __init__(self,            # [1] - Default constructor
           model_id: str,
           input_size: int,      # [2] - Size first layer
           output_size: int,    # [3] - Size output layer
           de_conv_modules: torch.nn.Sequential):
    super(DeConvModel, self).__init__(model_id)
    self.input_size = input_size
    self.output_size = output_size
    self.de_conv_modules = de_conv_modules


  @classmethod
  def build(cls,
      model_id: str,
      conv_neural_blocks: list,  # [4] - Input to the builder
      in_channels: int,
      out_channels: int = None,
      last_block_activation: torch.nn.Module = None) -> NeuralModel:
    
    de_conv_neural_blocks = []

    # [5] - Need to reverse the order of convolutional neural blocks
    list.reverse(conv_neural_blocks)

    # [6] - Traverse the list of convolutional neural blocks
    for idx in range(len(conv_neural_blocks)):
       conv_neural_block = conv_neural_blocks[idx]
       new_in_channels = None
       activation = None
       last_out_channels = None

        # [7] - Update num. input channels for the first 
        # de-convolutional layer
       if idx == 0:
            new_in_channels = in_channels
        
        # [8] - Defined, if necessary the activation 
        # function for the last layer
       elif idx == len(conv_neural_blocks) - 1:
          if last_block_activation is not None:
             activation = last_block_activation
          if out_channels is not None:
             last_out_channels = out_channels

        # [9] - Apply transposition to the convolutional block
      de_conv_neural_block = DeConvNeuralBlock.auto_build(
           conv_neural_block,
           new_in_channels,
           last_out_channels,
            activation)
      de_conv_neural_blocks.append(de_conv_neural_block)
        
       # [10]- Instantiate the Deconvolutional network 
       # from its neural blocks
   de_conv_model = DeConvModel.assemble(
       model_id, 
       de_conv_neural_blocks)
     
   del de_conv_neural_blocks
   return de_conv_model

The alternative constructor named build is designed to generate and set up the de-convolutional model using the convolutional blocks, referred to as conv_neural_blocks [#4].

To align the order of de-convolutional layers correctly, it's necessary to reverse the sequence of convolutional blocks [#5]. For every block within the convolutional network [#6], this method adjusts the number of input channels to match the number of input channels in the first layer [#7].

It then updates the activation function for the final output layer [#8] and systematically integrates the de-convolutional blocks [#9]. Ultimately, the de-convolutional neural network is composed using these blocks [#10]..

@classmethod
def assemble(cls, model_id: str, de_conv_neural_blocks: list):
   input_size = de_conv_neural_blocks[0].in_channels
   output_size = de_conv_neural_blocks[len(de_conv_neural_blocks)-1].out_channels 

   # [11]- Generate the PyTorch convolutional modules used by the default constructor
  conv_modules = tuple([conv_module for conv_block in de_conv_neural_blocks
                        for conv_module in conv_block.modules 
                        if conv_module is not None])
  de_conv_model = torch.nn.Sequential(*conv_modules)

  return cls(model_id, input_size, output_size, de_conv_model)

The assemble method is responsible for building the complete de-convolutional neural network. It does this by compiling the PyTorch modules from each of the blocks in de_conv_neural_blocks into a cohesive unit [#11].

Thank you for reading this article. For more information ...

References
[1] Tutorial: DCGAN in PyTorch
[2] A Gentle Introduction to Generative Adversarial Networks
[3] Deep learning Chap 9 Convolutional networks. 
          I. Goodfellow, Y. Bengio, A. Courville - 2017 MIT Press Cambridge MA 
[4] PyTorch
Source code: GitHub Neural Architecture


---------------------------
Patrick Nicolas has over 25 years of experience in software and data engineering, architecture design and end-to-end deployment and support with extensive knowledge in machine learning. 
He has been director of data engineering at Aideo Technologies since 2017 and he is the author of "Scala for Machine Learning" Packt Publishing ISBN 978-1-78712-238-3