In 2020, augmented reality will continue to take off like never before. While all kinds of 3D objects will have animated notifications and a new glasses technology will allow you to watch videos with voice and visis like Harry Potter, Google plans to use AI to enhance the moviegoing experience.
Recently, Google released a prototype of colorizing old YouTube videos, and a new paper from the research team shows how machine learning and deep learning can be used to improve colors. The paper is being published in ‘AI & Learning’, a research journal by Facebook.
The research team used a natural language processing (NLP) model trained on YouTube videos, their specific language, and objects. These videos were converted to video format by machine learning, meaning they are all pattern-recognizing videos. While previously machine learning models were used to improve the aesthetics of objects, this is the first time for a similar use case with visual media.
The test videos was taken from the ‘Creative Madness’ channel by Washington, DC-based blogger MySpace Haze.
To watch the video, just click below. I guess there is some kind of A/B testing going on here. However, the results are, well, stunning.
And here’s what it looks like using the improved colorization:
As you can see, in the original video, most of the objects have natural color hues. The new model, on the other hand, turns them into radiant, bright colors.
The research group has detailed how the NLP model works in the paper. Its goal is to create a colorizer that will “allow children to experience beauty, just like we experience beauty through art.”
This research is very cool, but there are some very good-natured and self-interested reasons to believe the project won’t exist in a few years’ time.
First of all, it would be great to have a machine learning model to colorize YouTube videos that you share on your social media profile. This is an excellent educational tool as it introduces people to the art and disciplines of color processing.
But the problem is that we all collect YouTube videos and videos on YouTube in our shared library that we upload to Youtube. While many of the videos may appear on this list, I wouldn’t be surprised if there are videos you didn’t catch just yet because you’re sharing them on your social network. YouTube doesn’t provide a download option for videos on its website. So if you share videos on your Facebook or Instagram with NLP enabled in your browser’s NLP panel, chances are you won’t be able to access the video file. YouTube doesn’t make a feature for NLP creation or storage. It’s going to be even more difficult if the researchers come up with a NLP model that works on hardware that uses deep learning too.
If you would like to keep reading about machine learning and AI, join our Machine Learning and AI speaker series.
Read next: How to use Caller ID to find and block spam calls and texts