There's something very special about black and white photography, isn't there? I'm not only referring to the nostalgic factor or the possibility of working with different parameters that highlight certain aspects and hide others. Each black and white photo carries a hint of mystery, and the most evident one is undoubtedly color. For decades, imaging and video experts have dedicated themselves to coloring frame by frame with quality and precision, but now, Algorithmia's Colorize system seeks to do that through artificial intelligence.
If my memory serves me right, I think most digital cameras on the market have a "black and white" mode to imitate the results of their "grandmothers", but many photographers out there think it's "not the same", and take a much more artisanal path. What brings us together here today is probably something equally complex, and it's the process of adding color to black and white photos, perhaps as a personal project, or as a major restoration. Of course, this is not exclusive to individual images. The world of cinema has witnessed spectacular restorations, like the case of "The Wizard of Oz" with Judy Garland, a film on which they worked frame by frame with the mission of assigning color, aligning, and eliminating imperfections. The hand and eye of a human are still needed to achieve such a level of detail, but the question is: How will artificial intelligences fare in the future?
The folks at Algorithmia have begun to answer (in their own way) with an online system we'll simply call Colorize. There's no technical information about how it works, but the basic description speaks of "deep learning" to apply color automatically. Colorize offers several sample images that let us appreciate its ability in a couple of seconds, although the key here is uploading our own images via a URL and seeing how it does. Colorize seems to have a more advanced command of landscapes, however, it's evident that the algorithm developed a certain preference for red and other similar tones.
In short, Colorize is still far from delivering professional results, and maybe that's not its goal, but when we talk about artificial intelligence, machine learning, and neural networks, time always plays in favor of those technologies. With each processed image, the system gains something extra, and in a couple of years, who knows?
