For years, people like Mark Truelove of Canadian Colour have been painstakingly colourizing archival photos using programs like Photoshop, putting in hours of work and research into each image they colourize. In that time, programmers have been been quietly honing algorithms that can produce the same type of work in seconds.
One such algorithm is called DeOldify, which was created by deep learning experts Jason Antic and Dana Kelley, in San Diego. It was built over two years, taking cues from millions of images.
We took 13 images from the City of Vancouver Archives holdings and ran them through a similar colourizing app, and were met with mixed results. Some of the photos looked as if they hadn't been colourized at all, some were very obviously the wrong colours, yet others looked pretty decent.
The ones you'll see in this gallery show the better samples. The quality is a far cry from images that are retouched by humans, but are still able to offer a different lens into our city's past than their black and white versions could.
.png;w=120;h=80;mode=crop)
