MSU Researchers Teach AI to Enhance Streaming Video on the Fly
Researchers at Moscow State University’s Artificial Intelligence Research Center have developed a way to increase the resolution and sharpness of streaming video without sacrificing processing speed.
The research was presented at ICLR 2026 in Rio de Janeiro. The problem is familiar to anyone who watches video online: when a connection is weak, streaming services compress the image, causing blur and block artifacts and erasing fine details. AI methods can increase resolution, but they are typically trained on data that bears little resemblance to real-world streaming video.
The researchers assembled a dataset of 5,200 videos spanning different genres and used it to compare 11 real-time enhancement methods. They also conducted a survey in which people chose which videos they thought looked better. Based on the results, the researchers proposed a model that runs fast enough for real-time use while doing a better job of restoring details.
The results could be used to develop services that enhance video and assess its visual quality.
