2020 is likely to be the first year for the application of AI technology in the field of media compression to be seriously considered. Here's a quick look at the four events that took place this year.
(1) Deep learning technology is missing from next-generation video standards
First, in July, the H.266/VVC standard was completed, followed by H.265/HEVC. In the standardization process, deep learning-based technologies were proposed in various fields such as intra block prediction and in-loop filtering, but in the end, they were left out of the standard, which made them regret. The main reason is that the performance versus complexity isn't enough. Is it still premature at this time? There were many stories to say.
(2) The emergence of MPAI, an AI-based media compression standardization organization
Second, MPEG, which has been leading the video compression standard, has been absorbed by a higher level organization and virtually shut down. Leonardo Chiariglione, the founder of MPEG, established a new non-commercial organization called MPAI (Moving Picture, Audio and Data Coding by Artificial Intelligence), and started a new attempt to standardize AI-based compression technology. The fact that the head of MPEG, who had been leading MPEG for decades, chose AI-based technology as the next step in itself became a hot topic, and it contrasted oddly with the decision to lack deep learning technology in H.266/VVC.
(3) AI-based video communication, NVidia Maxine announced
In October, NVidia unveiled a cloud-based video conferencing platform called Maxine. Compared to existing video conferencing systems using image compression technologies such as H.264/AVC, it is a technology that extracts facial landmarks, transmits them, and creates them in the terminal. By applying, we have announced the beginning of a war between the existing signal processing method and AI-based method in the video compression market.
(4) JPEG officially launched a new AI-based standard project
JPEG, an image standardization organization, conducted a call-for-evidence this year, a step to verify how significant performance improvement of AI-based image compression technology compared to the current best standard technologies (JPEG-2000, HEIF). A total of 4 proposals were submitted. As a result of the 89th JPEG meeting in October, the performance of AI-based image compression technology was remarkably excellent, and a new official project was decided to be launched. (Share the link below)
For reference, in order to create a standard, a project proposal, call-for-evidence (performance improvement verification), call-for-proposal (choosing a starting technology), and core experiment (partial technology optimization) are generally required. Unlike call-for-evidence, where anyone needs good performance, call-for-proposal is a process of choosing the first place among a number of proposals. It means that the war of the relevant technologies and their patents has just begun.
Various AI-based image compression technologies have been announced so far, and I wonder what combination of technologies will be the best in terms of performance and practicality. I am also wondering if JPEG, which has been applied to billions of units around the world, can be replaced by a new standard. I will continue to follow up this topic.
