Tensorflow implementation of the iColor user guided image colorization network.
- Linux
- Anaconda
- Install the required packages by running
setup.sh
.
./setup.sh
- Activate conda environment.
source activate icolor_tf
- Run
train.sh
to train the model. - Edit the
train.sh
to set the data source and appropriate batch size.- The image list should be a file containing a path to an image in each line, and the path will be concatenated with the image root to be the full path for accessing images.
- On a Titan X with 12GiB memory, use batch size 24.
- On a Tesla P100 with 16GiB memory, use batch size 32.
- The training output will be stored in the
output
directory, unless otherwise specified.