Computational Creativity with Neural Networks on images and videos

• 📖 Overview
  • Model architecture
  • 🛠️ Tech stack
  • 🚀 Features
  • � Results & Experiments
  • 🖼️ Screenshots & Examples
    • 1. Style Transfer comparison
    • 2. Style Transfer strength comparison
    • 3. Application screenshot - image style transfer
    • 4. Application screenshot - video style transfer
  • ⚙️ Installation
  • 📂 Project Structure
  • 🔮 Future Work
  • 👨‍🎓 Authors
  • 📄 License

📖 Overview

Neuro-Art is a research and experimental application developed as part of the engineering thesis "Computational Creativity with Neural Networks".
The project explores how artificial neural networks can be applied in computational creativity, with particular focus on:

• Artistic style transfer (for both images and video),
• Image morphing (smooth transitions between two images),
• Experiments with Variational Autoencoders (VAE) compared to classical style transfer models.

The application provides a graphical user interface for testing generative AI models and serves as a creative tool for designers and digital artists.

Model architecture

🛠️ Tech stack

• Machine Learning: PyTorch Lightning, Tensorflow
• GUI: PyQt6
• Image Processing: OpenCV, Pillow, NumPy
• Hardware Acceleration: CUDA (NVIDIA GPU support)

🚀 Features

• Image Style Transfer – transform a photo into the style of a famous artwork.
• Video Style Transfer – apply artistic styles to entire video sequences.
• Morphing – generate smooth transitions between two input images.
• VAE-based Transfer – alternative style transfer using latent space interpolation.

📊 Results & Experiments

• Comparison of classical vs. VAE-based style transfer,
• Evaluation of morphing algorithms (traditional vs. neural),
• Video style transfer performance improvements using interpolation,
• Many visual examples of generated artistic outputs.

🖼️ Screenshots & Examples

1. Style Transfer comparison

2. Style Transfer strength comparison

3. Application screenshot - image style transfer

4. Application screenshot - video style transfer

(All examples were generated using the implemented system.)

⚙️ Installation

1. Clone the repository:

   git clone https://github.com/neuroviscode/neuro-art.git
   cd neuro-art

2. Create a virtual environment and install dependencies:

   python -m venv venv
   source venv/bin/activate   # Linux/MacOS
   venv\Scripts\activate      # Windows
   
   pip install -r requirements.txt

3. Install CUDA for GPU acceleration.

4. Run the application:

   python main.py

📂 Project Structure

neuro-art/
│── logic/              # Core logic and ML models
│   ├── morphing/       # Morphing algorithms
│   ├── preprocessing/  # Image loading & preprocessing
│   ├── style_transfer/ # Classical style transfer
│   ├── style_transfer_vae/ # VAE-based style transfer
│   └── vae_models/     # VAE architectures
│
│── widgets/            # GUI (PyQt6)
│   ├── home/           # Home screen
│   ├── library/        # User's saved works
│   ├── morphing/       # Morphing interface
│   ├── style_image/    # Image style transfer
│   ├── style_video/    # Video style transfer
│   └── settings/       # Application settings
│
│── assets/             # Resources
│   ├── examples/       # Example input images
│   ├── icons/          # UI icons
│   ├── models/         # Pretrained models
│   └── results/        # Generated outputs

🔮 Future Work

• Integration with diffusion models (Stable Diffusion, DALL·E, MidJourney-like approaches),
• Latent-space morphing for smoother artistic transitions,
• Web-based version with API support,
• Cloud deployment for GPU rendering.

👨‍🎓 Authors

Developed as part of the Engineering Thesis
Computer Science – Intelligent Interactive Systems
Gdańsk University of Technology

• inż. Paweł Cichowski
• inż. Michał Cellmer
• inż. Jakub Link

Supervisor: dr hab. inż. Julian Szymański

📄 License

This project is released under the MIT License.
See LICENSE for details.