AI for Video Calls

Streaming raw digital video requires large amounts of memory, so smartphones and computers generally compress the data by cutting extraneous information from each video frame. At slower speeds, this can cause the images to become pixelated and erratic. But a neural network was recently tested that uses less memory by essentially “guessing” what much of the picture should look like and replace portions of it with objects from its own database.

For example, the AI system can recognize a tree from its own library of textures and features. So if an image contains a tree, it deletes that portion from the original image and inserts its own picture of a tree. In a test conducted on 180 subjects, the same images were compressed using the AI algorithm and a standard compression method. Over 80 percent of participants preferred the AI version at very low data rates.