Implementation of some automatic colorization models using deep neural network:
- Implementation of the regression-based model provided in: "Let there be Color!: Joint End-to-end Learning of Global and Local Image Priors for Automatic Image Colorization with Simultaneous Classification" Link to the original paper
- Implementation of a classification-based model inspired in part by Zhang et al. in "Colorful Image Colorization" [Link to the original paper] and R.Dah in here
- Implementation of a regression-based model inspired from this Medium blog article
You can consult the project report here (in French)
This project runs on python >= 3.6, use pip to install dependencies:
pip3 install -r requirements.txt
Use the main.py
script to choose the model to train and the parameters to use
usage: main.py [-h] [-m {naive,reg,classif}] [-d DATA_DIR] [-e EPOCHS]
[-b BATCH_SIZE] [-tc TO_COLOR] [--no-load] [--no-save]
[--no-train] [--early]
An implementation of multiple approachs to automatically colorize grey-scale
images
optional arguments:
-h, --help show this help message and exit
-m {naive,reg,classif}, --model {naive,reg,classif}
Colorization model
-d DATA_DIR, --data-dir DATA_DIR
Directory where data is
-e EPOCHS, --epochs EPOCHS
Number of epochs to train for
-b BATCH_SIZE, --batch-size BATCH_SIZE
Batch size to use
-tc TO_COLOR, --to-color TO_COLOR
Number of samples to be colored
--no-load Disable loading saved model
--no-save Disable saving the new model
--no-train Disable training the model
--early Enable early stopping
Use the levi.py
script to color images using a model
usage: levi.py -h [-d DATA_DIR]
An implementation of multiple approachs to automatically colorize grey-scale
images
optional arguments:
-h, --help show this help message and exit
-m {naive,reg,classif}, --model {naive,reg,classif}
Colorization model
-d DATA_DIR, --data-dir DATA_DIR
Directory where the images to be colored is
This project is under development. The final objective being to create a model that colors mangas by learning on the corresponding anime, this project will be updated regularly.