Tuesday, May 22, 2018

High-Dynamic Range and Artificial Intelligence Stand Out in Display Week Monday Seminars

Monday, May 21, saw the continuation of Display Week educational opportunities, as experts in various fields shared from the fruits of their expertise in a series of 90-minute seminars. Two notable examples stood out. One was given by Dolby Laboratories engineers Timo Kunkel and Rob Wanat, who gave a very good seminar titled “High-Dynamic-Range: A Consumer Ecosystem.” The second was another excellent offering from Achin Bhowmik called “Artificial Intelligence: Image Recognition and Visual Understanding.”

Yes, I know that Dolby has a point of view regarding HDR. And yes, it has a proprietary scheme for handling HDR content as part of its licensing business model. But its seminars and papers are always well presented and backed up by good research done by knowledgeable engineers. The speakers energized their talk with a description of how the human visual system sees the world and responds to high-dynamic range scenes in the real world. This informs its approach of how to best acquire, deliver, and display HDR content. Kunkel and Wanat built a good case for the Dolby Perceptual Quantizer as embodied in SMPTE 2084 as a good way to encode EOTF for HDR displays. It will be interesting to see how HDR continues to evolve as displays improve and HDR standards and pipelines are developed.

Achin Bhowmik delivered another great talk with his primer on AI as applied to image recognition. He gave a brief history of AI and the breakthroughs that resulted in image-recognition performance that surpassed that of humans a few years ago. This was enabled by using programming techniques inspired by human brain physiology, advances in computing, and the availability of lots of image data for training the algorithms. He gave us a basic understanding of how “deep” neural networks are trained with some simple examples. In addition to recognizing image content, neural networks can also be trained to achieve semantic scene understanding. See the “black and white dog jumps over bar” above, for an example of a scene recognized and described via AI.  There is a lot of potential for AI and many organizations are making some big bets on its continued development. There’s still a bit of work to do, however, before I would be comfortable letting AI drive my car or fly my plane. I do appreciate those email spam filters, though… -- Tom Fiske

No comments:

Post a Comment