Kara

Kara (Kolouring Artificial Assitant), is an AI bot the colourize B&W photos.

All the images used in the dataset are from Unplash and are creative common pictures by professional photographers. It includes 9.5 thousand training images and 500 validation images.

Usage

To retrain the model with the already available dataset, run:

cd kara/
python3 main.py createmodel

To load the already trained model (there already is a trained model in the project):

cd kara/
python3 main.py loadmodel

To predict specific images located in your filesystem, run:

cd kara/
python3 main.py f <filename1> <filename2>

Each <filename1> represents a path to a specific image in your system. Multiple images are allowed, just separate each path with spaces.

Dependencies

• Python 3.7
• Standard Version

Configuration

To be done.

Development

Installing VirtualEnvWrapper

We recommend using a virtual environment created by the virtualenvwrapper module. There is a virtual site with English instructions for installation that can be accessed here. But you can also follow these steps below for installing the environment:

sudo python3 -m pip install -U pip             # Update pip
sudo python3 -m pip install virtualenvwrapper  # Install virtualenvwrapper module

Observation: If you do not have administrator access on the machine remove sudo from the beginning of the command and add the flag --user to the end of the command.

Now configure your shell to use virtualenvwrapper by adding these two lines to your shell initialization file (e.g. .bashrc, .profile, etc.)

export WORKON_HOME=\$HOME/.virtualenvs
source /usr/local/bin/virtualenvwrapper.sh

If you want to add a specific project location (will automatically go to the project folder when the virtual environment is activated) just add a third line with the following export:

export PROJECT_HOME=/path/to/project

Run the shell startup file for the changes to take effect, for example:

source ~/.bashrc

Now create a virtual environment with the following command (entering the name you want for the environment), in this example I will use the name kara:

mkvirtualenv -p $(which python3) kara

To use it:

workon kara
sudo python3 -m pip install pipenv
pipenv install # Will install all of the project dependencies

Install tensorflow:

pip install tensorflow==1.14

Due to some error while locking packages stage, as can be seen in this issue, the tensorflow installation has to be manual.

Observaion: Again, if necessary, add the flag --user to make the pipenv package installation for the local user.

Local Execution

For local system execution, run the following command in the project root folder (assuming virtualenv is already active):

cd kara
python src/main.py createmodel

This will train and run the system on your machine. This way you can test new implementations or new optmizations. Also you can color the images for testing.

Test

Lint

To lint your code follow the script bellow:

1. Enable virtualenv color;

2. Ensure that the dependencies are installed, especially:

flake8

3. Run the command below:

cd kara/
flake8 src/

During the lint process the terminal will report a code errors and warnings from the PEP8 style guide, for more configurations and additional documentation go to flake8 and PEP8.

Build

Generate Changelog

To generate changelog we use the standard version tool, it will auto generate a new changelog for every new release by using the commit messages. To generate a new release and generate the updated changelog just do the following steps:

1. Install all dependencies

yarn install

2. Run standard version:

npm run release

If the release is a pre-release you should add the --prerelease to the command:

npm run release -- --prerelease alpha

For further instructions or other options check the full documentation of standard version project in the CLI Usage section.

Contributors

Project contributors

Name	Registration
João Pedro Sconetto	14/0145940
Victor Correia de Moura	15/0150792

References

Iizuka, Satoshi & Simo-Serra, Edgar & Ishikawa, Hiroshi. (2016). Let there be color!: joint end-to-end learning of global and local image priors for automatic image colorization with simultaneous classification. ACM Transactions on Graphics. 35. 1-11. 10.1145/2897824.2925974.

Wallner, Emil (Oct, 2017). Colorizing B&W Photos with Neural Networks. Floydhub. Access date: 7 Nov 2019. Available at: https://blog.floydhub.com/colorizing-b-w-photos-with-neural-networks/