Abstract

Automatic colorization is the process of adding color to greyscale images. Wecondition this process on language, allowing end users to manipulate acolorized image by feeding in different captions. We present two differentarchitectures for language-conditioned colorization, both of which produce moreaccurate and plausible colorizations than a language-agnostic version. Throughthis language-based framework, we can dramatically alter colorizations bymanipulating descriptive color words in captions.

 

Quick Read (beta)