Given a gray scale photograph as input, this approach creates a plausible color version of the photograph. We propose a fully automatic approach that produces vibrant and realistic colorization. We embrace the underlying uncertainty of the problem by posing it as a classification task and use class re-balancing at training time to increase the diversity of colors in the result. The system is implemented as a feed-forward pass in a CNN at test time and is trained on over a million color images. We evaluate our algorithm using a “colorization Turing test”, asking human participants to choose between a generated and ground truth color image. Our method successfully convince humans on 32 percent of the trials, significantly higher than previous methods. Moreover, we show that colorization can be a powerful pretext task for self-supervised feature learning, acting as a cross-channel encoder. This approach results in state-of-the-art performance on several feature learning benchmarks.

Given the lightness channel L, this system predicts the corresponding a and b color channels of the image in the CIE Lab color space. Here, L* represents lightness ranging from black(0) to white(255), a* from green(-) to red(+), and blue(-) to yellow(+).