This repository contains a project developed for the Deep Learning course during the 2024/2025 Summer Semester at the Warsaw University of Technology, Faculty of Mathematics and Information Science.
The dataset can be obtained from various sources, but we recommend using the fferlito/Cat-faces-dataset repository.
To download the dataset on Linux, run the following commands:
wget [https://github.com/fferlito/Cat-faces-dataset/raw/master/dataset-part1.tar.gz](https://github.com/fferlito/Cat-faces-dataset/raw/master/dataset-part1.tar.gz)
wget [https://github.com/fferlito/Cat-faces-dataset/raw/master/dataset-part2.tar.gz](https://github.com/fferlito/Cat-faces-dataset/raw/master/dataset-part2.tar.gz)
wget [https://github.com/fferlito/Cat-faces-dataset/raw/master/dataset-part3.tar.gz](https://github.com/fferlito/Cat-faces-dataset/raw/master/dataset-part3.tar.gz)Extract the downloaded archives into the cats/images directory:
mkdir -p cats/images
tar -xzvf dataset-part1.tar.gz -C cats/images --strip-components=1
tar -xzvf dataset-part2.tar.gz -C cats/images --strip-components=1
tar -xzvf dataset-part3.tar.gz -C cats/images --strip-components=1To set up the environment on Linux, create a virtual environment and install the required packages:
python3 -m venv venv
source venv/bin/activate
pip install torch torchvision torchsummary matplotlibThis project implements four different architectures for image generation:
- Autoencoders
- Variational Autoencoders (VAEs)
- Diffusion Models
- GAN
Each model architecture is maintained in a separate branch within this repository.
Below are samples of the generated images.
Note: Please note that all images generated using the standard Autoencoder exhibit a distinct blue artifact in the center. The root cause of this phenomenon requires further investigation.
