Artistic Style Transfer for Images and Videos

Welcome to our repository showcasing Python scripts that enable the application of artistic style transfer to both real-time webcam videos and pre-recorded videos/images. This implementation harnesses the power of TensorFlow, PyTorch, and OpenCV to create captivating visual transformations.

Example Stylized Image

Behold a captivating stylized image produced by our script:

Example Stylized GIF

Observe the mesmerizing GIF stylization in action:

NOTE: Our solution isn't limited to images and GIFs – it's also applicable to full-fledged videos in formats such as MP4 and MPEG4. For real-time stylization, you'll need to run the streamlit_app.py script on your localhost.

Getting Started

Follow these steps to set up and run the scripts on your local machine.

Prerequisites

Before you begin, ensure that you have the following software installed:

• Python (version 3.6 or higher)
• Git

Cloning the Repository

To acquire the repository on your local machine, open a terminal and execute the following command:

git clone https://github.com/qudus4l/Style-Transfer.git

Installing Dependencies

Navigate to the project directory:

cd Style-Transfer

Install the necessary Python packages using the pip package manager:

pip install -r requirements.txt

Running the Script

To initiate the real-time style transfer script using Streamlit, execute the following command:

streamlit run streamlit_app.py

Stopping the Script

During script execution, you can halt the real-time style transfer by closing the Streamlit app.

PyTorch Implementation

Curious about the inner workings of our project? We've harnessed the power of PyTorch to drive the artful transformation process. Dive into the PyTorch implementation in the pytorch_implementation/src/train.py script.

To explore the PyTorch implementation:

1. Navigate to the pytorch_implementation directory.
2. Open the src folder.
3. Check out the train.py script to uncover the PyTorch-driven artistic transformation process.

Remember, this implementation provides a peek into the fascinating world of style transfer and offers insights into how the art evolves throughout the creative journey.

Note: If you decide to run the PyTorch implementation locally, ensure you have the necessary dependencies installed as outlined in the main README. You can experiment with various settings, such as epochs and style weights, directly in the script. The stylized images are saved every 100 epochs to help you observe the artistic progression.

We recommend using a GPU for faster training, and if you're on an Apple Silicon device, you'll be pleased to know that support for MPS (Metal Performance Shaders) is available for accelerated computation.

Acknowledgments

Our implementation leverages a variety of technologies, including TensorFlow, TensorFlow Hub, PyTorch, and other libraries, to achieve real-time style transfer.

License

This project operates under the MIT License. Please refer to the LICENSE file for detailed information.