Nvidia Research develops a neural network to replace traditional video compression

What They Say

Nvidia has develoed a new cloud-AI-based codec for video which is claimed to reduce bandwidth by 90% compared to H.264. (Hmmm, that’s what the article said, but an example shown dropped the data rate from 97.28KB/frame to 0.1164KB/frame which is a 1000X reduction, although, of course, audio compression would not be affected) The algorithm is specific to video conferencing and sends keypoints of a face rather than the full image. The face image is reconstructed using the keypoint data. The concept can also be used to create cartoon characters that are animated and to solve the problems of eye contact. See the video below.

What We Think

It’s only a couple of weeks since I commented in a Display Daily on the eye contact issues of video conferencing and here is another solution to the problem. (Barriers to VideoConferencing, One Down….). As usual, I suspect this technology could also be used for ‘deep fakes’. (BR)

keypoints