Remember that amazing automatic colorizing algorithm we told you about back in March? It was just put to an interesting test. As a fun “what if?” hobbyist Amir Avni tried the neural network-powered colorizer on a B&W video.
It should be obvious that this isn’t a simple transition. To go from colorizing a still image by learning from thousands of other images online is one thing, to do it 24 times per second, smoothly, is entirely another.
“As a bonus, I took the bot’s algorithm and started testing it on some video-clips,” writes Avni. “Of course the results aren’t perfect since it was trained on photos and not on a batch of photos, but the results are still interesting.”
Interesting they are. The colors vary a ton, but with a fairly static clip like the Beatles video, the bot can do a decent job of staying consistent.
Problems arise when you get a more traditional music video with transitions and varying scenes, as you can see with this colorized version of Black’s Wonderful Life:
The research has a long way to go before “Content Aware Colorize” shows up in your Tools on Photoshop (or Premiere Pro), but the proof of concept is there. Not sure about you, but we can’t wait to see where this goes in the coming years!