Icon Generation with Conditional GANs

This repository contains the code for the paper Icon Generation with Conditional GANs by Diogo Medeiros.

Table of Contents

• Icon Generation with Conditional GANs
  • Table of Contents
  • Abstract
  • Usage
    • Requirements
    • Configuration
    • Training
    • Generation
  • Results
    • Metrics
    • Generated Icons
  • License
  • Acknowledgments
  • Cite this work

Abstract

Conditional Generative Adversarial Networks (cGANs) were used to generate icons with specific characteristics or properties given the Icons-50 dataset.

The performance of the cGANs was impacted by the imbalanced nature of the original dataset. Filtering the original dataset to include only the 10 most frequent classes (the Icons-10 dataset) improved the performance of the cGANs, allowing them to generate more detailed and striking icons.

This paper demonstrates the potential of cGANs for use in applications where a large number of icons with specific characteristics are needed.

Usage

Requirements

• Python 3.8+
• TensorFlow 2.9.x
• Hydra Core 1.1.x

Configuration

The configuration files are located in the config directory. The config.yaml file contains the default configuration for the project. For more information on how to use Hydra, please refer to the documentation.

The configuration file contains the following parameters:

Parameters for the model architecture and data preprocessing:

• latent_dim: The dimensionality of the latent space.
• image_size: The size of the images in the dataset.
• channels: The number of channels in the images in the dataset.
• num_classes: The number of classes in the dataset.
• batch_size: The batch size for training.
• epochs: The number of epochs for training.
• top_k: The number of classes to use in the dataset. If set to 0, all classes are used.
• shuffle: Whether to shuffle the dataset.

Parameters for the training process:

• lr: The learning rate for the Adam optimizer.
• beta_1: The beta_1 parameter for the Adam optimizer.

Parameters for the model and dataset paths:

• data_path: The path to the dataset.
• classes_path: The path to the file containing the class names.
• save_path: The path to save the model checkpoints.
• filt_save_path: The path to save the model checkpoints for the filtered dataset.

Training

To train a model, run the following command:

python train.py

Generation

To generate icons, run the following command:

python test.py

Results

Metrics

The following metrics were used to evaluate the performance of the cGANs:

• Binary Cross-Entropy: The binary cross-entropy loss function used to train the cGANs.
• Binary Accuracy: The binary accuracy metric used to train the discriminator.

Loss	Accuracy

Generated Icons

The following icons were generated by the cGANs:

Class	Icons-50	Icons-10
Clock
Family

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

This project was developed as part of the class Deep Learning at UTAD, in the context of the Mestrado em Engenharia Informática.

Cite this work

@misc{Medeiros2023Icon,
    title = {Icon Generation with Conditional GANs},
    author = {Diogo Medeiros},
    year = {2023},
}