YouTuber uses neural networks to upscale 1896 short film to 4K 60 fps

0

There are many times when we see ‘future technology’ that just feels so out of place. We call it a marketing gimmick and move on. However, a YouTuber has shown us that not everything is unreal. There are some technological advancements that require using the adage “so good to be true.”

A YouTuber, Denis Shiryaev, has used the technological advancements available and turned a black and white movie from the year 1896 into a movie with 4k crystal clarity, running at 60 frames per second.

We have all been so numbed to the technological buzzword of Artificial Intelligence, claimed to be used by every tech firm out there. Artificial intelligence has had a long history with humans and software development.

The idea kicked off with the objective of understanding and imitating how a human brain learns. Coupled with this are the concepts of machine learning, deep learning, neural networking, etc. While they all probably sound like a gimmick, just words without meaning, it is not so. There is actual scientific technology behind this.

These terms indicate the ideas being used by a service to deliver to the end-consumer a smart product in the form of consumer electronics. It is also quite true that artificial intelligence is being used almost everywhere gadgets are present. You can see the examples in how Instagram is using it to identify and fight abuse on its platform, in how YouTube recommends videos to you.

In fact, it has already been so integrated in our lives that we do not notice it anymore. Neural Network is a bit more complex, where layers of interconnected nodes are trained to recognize patterns by supplying the network with a large amount of data and then filtering the parameters.

How Did He Achieve This?

The original video of the clip he used was only 640 by 480 resolution and 20 frames per second. This does not mean much today but was a common back in the day. Denis, to increase the quality and frame rate, used a mix of neural networks from a product called Gigapixel AI, and a technique called depth-aware video frame interpolation. What this does is not only does it upscale the resolution of the video (in this case to 4k) but helps better the frame rate as well.

Gigapixel AI is an upscaling process developed by Topaz Labs to be used for increasing the resolution of media. The AI does so by filling in the information in the given media by using patterns and structures from a large pool of source images that are downscaled so that the image can be comparable to the originals.

The AI here does not rely on interpolation like standard techniques while adding in mixing pixels. However, Topaz Labs itself has said that the process is much slower than the standard techniques of bi-cubic, Lanczos, spline, etc and it might also introduce erroneous details.

Read – Facebook Secretly Built a Facial Recognition App That Let Employees Identify People by Pointing a Phone at Them

But all that aside, the biggest takeaway is that neural networks can be trained to do a variety of tasks including colorizing images and adding in missing textures or patterns. In fact, we have seen the potential of neural networks when it was used to overhaul the video game Doom to give it a higher quality and its characters more convincing animations. It is a tool of immense possibility and will most certainly change how we use technology in the near future.

LEAVE A REPLY

Please enter your comment!
Please enter your name here