📊🔬🤖🚀
Welcome to the GANalytics repository! This repository contains the codebase for StyleGAN Model with pytorch implementation. This codebase is part of a larger project that conducts a detailed analysis of the performance comparison between various generative models. Please see the presentation video here.
GANalytics/
│
│── dataset/
│ └── Flowers (to be downloaded)
│ └── Pokemon (to be downloaded)
│
│── README.md
│── requirements.txt
│── dataset_utils.py
│── config.py
│── model.py
│── train_stylegan_model.py
│── calculate_scores.py
│── visualize_results.ipynb
This project focuses on the implementation of StyleGan implemented in pytorch. We provide implementations for various components, including dataset preparation, model training, and inference.
To run the code in this repository, you'll need to install the dependencies specified in requirements.txt:
You can install the required packages using pip:
pip install -r requirements.txt
To use the code in this repository, follow these steps:
Include the all the training images under a subdirectory in the dataset directory.
Modify config.py to add or edit dataset configurations, such as dataset location, and dataset name.
Adjust the hyperparameters as needed at the top of train_stylegan_model.py as needed.
To train the model, run the following command.
python train_stylegan_model.py {dataset name}.
I have trained this model on the oxford flowers dataset and a subset of the Pokemon dataset.
When the model is trained, parameters checkpoint will be saved to ./checkpoint/trained.pth every 1000 iterations.
Fréchet Inception Distance is used to estimate the performance. Please see calculate_scores.py for more details.
Please use plot_results.ipynb for visualizing training results
A figure of the StyleGAN model architecture is shown below.
Sample images generated from training the StyleGAN model on the Oxford Flowers dataset are shown in the following figure.
Implementing style-mixing feature is considered as a next step.
https://www.youtube.com/watch?v=yZ9bIT23XEw
Contributions to this project are welcome! If you'd like to contribute, please follow these guidelines:
Fork the repository. Create a new branch for your feature or bug fix. Make your changes and commit them with descriptive messages. Push your changes to your fork. Submit a pull request to the main repository. Feel free to reach out at shaoyuli@alumni.cmu.edu for any questions, discussions, or collaboration opportunities!