image-colorization

Simple image colorization with Pytorch to colorize black-and-white landscape images
A convolutional neural network is trained with 800 grayscale landscape images to perform image colorization on gray images.

Usage

$python3 color_img.py input-image-path output-image-path
or
$python3 -W ignore color_img.py input-image-path output-image-path
(-W ignore can be used to avoid warning messages)

Example

$cd src
$python3 -W ignore color_img.py ../test.jpg ../out.jpg
-> Result image is saved to ../out.jpg

Above two images are showing the performance of the model on a given random test image.
Image on the left is a grayscale input image.
Image on the right is the colorized output version.

Whereas the groundtruth image(64x64) looks like this:

Still pretty good, huh :)

P.S. Model performs well on grayscale images of landscapes because almost all of the training data used are grayscale landscape images that you can see from the data folder with their groundtruth images (color64 folder).