Neural Style Transfer with VGG-19

A Python implementation of neural style transfer using two distinct approaches to compare performance and output quality.

Overview

This repository implements neural style transfer algorithms using two different methodologies:

1. Custom Implementation: Building the algorithm from scratch using TensorFlow's linear algebra functions, following Gatys et al.'s original approach with VGG-19 convolutional layers
2. Pre-trained Model: Using TensorFlow Hub's pre-trained neural artistic stylization network proposed by Ghiasi et al.

The project compares the outputs and performance of both implementations to evaluate their effectiveness.

Sample Output

Project Structure

├── images/
│   ├── content/          # Content images for style transfer
│   ├── style/            # Style reference images
│   └── stylized_output/  # Generated stylized images
├── notebooks/
│   └── Neural_Style_Transfer_Base_Implementation.ipynb
└── README.md

Getting Started

Prerequisites

• Python 3.7+
• TensorFlow 2.x
• Jupyter Notebook
• GPU support recommended for faster processing

Installation

1. Clone the repository:

git clone https://github.com/your-username/Neural_Style_Transfer_VGG19.git
cd Neural_Style_Transfer_VGG19

2. Install required dependencies:

pip install tensorflow tensorflow-hub jupyter matplotlib numpy pillow

Docker Usage

Note: Currently, no Dockerfile is present in this repository. To run this project with Docker, you would need to:

1. Create a Dockerfile with the appropriate Python environment and dependencies
2. Build the Docker image
3. Run the container with volume mounts for the images directory

If you have a Dockerfile, you can run:

# Build the image
docker build -t neural-style-transfer .

# Run the container
docker run -it -v $(pwd)/images:/app/images neural-style-transfer

Usage

1. Open the Jupyter notebook:

jupyter notebook notebooks/Neural_Style_Transfer_Base_Implementation.ipynb

2. Follow the step-by-step instructions in the notebook

3. Explore the generated stylized images in the /images/stylized_output/ folder

Additional Resources

• Blog Post - Detailed implementation walkthrough
• Open in Colab - Run directly in Google Colab