PixColor: Pixel Recursive Colorization

TL;DR

PixColor introduces a two-stage neural network approach that generates diverse, high-quality colorizations of grayscale images by first creating a low-resolution color version and then refining it to high resolution.

Contribution

The paper presents a novel two-step CNN-based method for automatic image colorization that improves diversity and plausibility over previous techniques.

Findings

Produces more diverse colorizations than existing methods

Achieves higher plausibility as judged by human raters

Effectively combines low-res and high-res colorization stages

Abstract

We propose a novel approach to automatically produce multiple colorized versions of a grayscale image. Our method results from the observation that the task of automated colorization is relatively easy given a low-resolution version of the color image. We first train a conditional PixelCNN to generate a low resolution color for a given grayscale image. Then, given the generated low-resolution color image and the original grayscale image as inputs, we train a second CNN to generate a high-resolution colorization of an image. We demonstrate that our approach produces more diverse and plausible colorizations than existing methods, as judged by human raters in a "Visual Turing Test".