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

Summary: The process these researchers propose, Space-Time Video Super-Resolution (STVSR), uses a single processing step instead of two steps like other methods in use today (VFI networks).

This new process inserts “missing” video frames based on the existing frames and then stitches them together simultaneously. While that’s happening, the AI network predicts slow-motion video frames and inserts them into the video as well.

What They Said: The researchers conducted several experiments to see if their model would produce better and faster results than current VFI networks. Their system has shown a large improvement in the processing speed and size required for the AI network so far.