AI-assisted and AI-based Image/Video Processors and Codec

AI has been used almost everywhere. Video analytics is the most commonly used AI application in conjunction with image/video. In the era of Internet of Video Things (IoVT), many cameras collect a massive amount of visual data to be analyzed by AI. While there are still challenges in the algorithms, especially after the systems are deployed, e.g., accuracy, energy efficiency, and processing speed, we can further use AI to design efficient IoVT systems at the development time. This talk will discuss the following two topics:

1. Deployment-time AI for Image/Video Processors and Codec
2. Development-time AI for Image/Video Processors and Codec

In deployment-time AI, we use AI to achieve system function (e.g., analyzing video, detecting/recognizing objects in the image/video) after the hardware system has been designed.

In development-time AI, we use AI to help us design AI better. For example, EDA tools and compilers use AI to help people generate better designs. There are many design tradeoffs in image/video processing/compression. At the development time, using AI (instead of using human hand-craft heuristics) can help us design the hardware system better, even before the system is developed.