UVQ is a no-reference perceptual video-quality assessment model developed by Google Research / YouTube.

“No-reference” means that you don’t have to provide a “reference” or “original” video to compare the content against — the model works with any live content. “Perceptual” means that the model tries to give a rating that would match a human’s perception of the video quality.

UVQ returns a “Mean Opinion Score” (MOS), which is a number between 1 and 5:

The Stb-tester Node has a decent GPU, so our UVQ implementation runs in near real-time, allowing you to sample about 1 second of video out of every 5 seconds.

To learn more:

• UVQ: Measuring YouTube’s Perceptual Video Quality on the Google Research blog has high-level information about the algorithm.
• stbt.UVQ API documentation in the Stb-tester manual.