Learning Representations for Automatic Colorization

ECCV 2016

Successful examples of grayscale images that were automatically colorized by our algorithm:

Browse all 10,000 colorized images here.


We develop a fully automatic image colorization system. Our approach leverages recent advances in deep networks, exploiting both low-level and semantic representations during colorization. As many scene elements naturally appear according to multimodal color distributions, we train our model to predict per-pixel color histograms. This intermediate output can be used to automatically generate a color image, or further manipulated prior to image formation; our experiments consider both scenarios. On both fully and partially automatic colorization tasks, our system significantly outperforms all existing methods.

Overview of network architecture for automatic colorization

Our automatic colorizer takes each pixel, looks at a its surrounding and predicts a distribution over plausible colors.


You can install the colorizer via pip (make sure you have Caffe with Python bindings installed):

pip install autocolorize

This will install the command autocolorize:

autocolorize grayscale.jpg -o colorized.jpg

There is a start-up cost to loading in the model, so to colorize a batch, use:

autocolorize *.jpg -o output

The colorized images will be saved to the output directory.


Related work

Automatic colorization has gained a lot of interest recently. In particular, Zhang et al. and Iizuka & Simo-Serra et al. who concurrently developed a colorization system.