Researchers Create Improved Video-Resolution Enhancement AI

It's both 3 times faster and uses a network that's 4 times smaller

by
Rob LeFebvre
Rob LeFebvre
Senior News Editor
Rob LeFebvre has been a freelance technology writer for 10 years and an educator for 20. His articles have appeared in 148Apps, Cult of Mac, Engadget, and many others.
our editorial process
Rob LeFebvre
Published February 27, 2020
Updated February 27, 2020 01:44PM EST

Why This Matters

Enhancing video will continue to improve, having a big impact in entertainment, law enforcement, and consumer video. Making the AI network involved faster and smaller in size will help bring future video enhancement tools to all of us, possibly in our own personal devices.

Grid of video frames being improved by AI
Purdue, Rochester, Northeastern Universities

Dramatically improving the resolution of video has been in the spotlight lately, with incredible transformations of old film into modern 4K resolution. The technology to do so is rapidly improving, as well. Researchers at Purdue University, the University of Rochester, and Northeastern University have come up with a way to drastically increase the resolution of video at three times the speed of current methods using an AI network that is four times smaller.

Extensive experiments show that our one-stage framework is more effective yet efficient than existing two-stage networks.

State of the Art: The process these researchers are proposing, Space-Time Video Super-Resolution (STVSR), uses a single-stage processing pass instead of two stages, like other methods in use today (called VFI networks).

This new process inserts "missing" video frames based on the existing frames, then aggregates them at the same time. While that's going on, the AI network predicts slow-motion video frames and places them into the video as well.

What they said: The researchers performed several experiments to see if their model would produce better and faster results than current VFI networks. Their system has thus far shown a large improvement in speed of processing and the size the AI network needed.

Why do you care: As the ability to drastically improve the resolution of older video and film continues to get faster and take up less space, it's easy to imagine a near-future system residing on your own laptop or smartphone. Imagine all your home movies from the days of standard definition, documents both personal and useful to history, getting the 4K treatment. Think of documentaries bringing new light and insight to old film documents, and potential improvements to law enforcement and surveillance footage. Someday soon, the Hollywood "enhance" button may be a real thing, and it might even be on your iPhone.

Via: VentureBeat

Become an Expert on Video Resolution