tomasmikeska / deep-colorization

Automatic grayscale image colorization using deep learning

Deep colorization

Implementation of deep learning model in Keras for image colorization. Project uses U-Net trained as Self-Attention GAN together with Perceptual loss instead of usual MAE or MSE. Work is still in progress.

---

Latest Examples

Historical photos

"The Roaring Lion", Winston Churchill's iconic portrait, 1941

Lower Manhattan’s Classic Skyline Seen Aerially From Battery Park, 1956

"Migrant Mother" by Dorothea Lange, 1936

NYC street vintage photo

Metropolis (movie), 1927

Original RGB images (left) converted to grayscale and colorized (right)

---

Requirements

• Python 3.x
• pip

Installation and setup

Install pip packages using

$ pip install -r requirements.txt

Add .env file to project root with environmental variables

COMET_PROJECTNAME={comet_project_name}
COMET_WORKSPACE={comet_workspace}
COMET_API_KEY={comet_api_key}

[optional]

There is a Docker image included that was used for training in cloud. You can build it from local Dockerfile with

docker build -t ml-box .

or get it from Docker Hub

docker pull tomikeska/ml-box

Usage

Train model using command

$ python src/train_gan.py

Colorize image using trained weights

$ python src/colorize.py --weights model/weights.h5 --source source.jpg --output output.jpg

License

Code is released under the MIT License. Please see the LICENSE file for details.