CVPR Daily - 2018 - Thursday

John R. Smith is a researcher at IBM and an IBM Fellow. He manages a research team at IBM T. J. Watson Research Center working on computer vision, language, speech technologies and knowledge interaction. IBM has come in force to CVPR this year. Yes, CVPR is a big conference for us. It is one of the big conferences in AI actually. We can see this year the attendance is tremendous. I think more than 6,000 people! 6,500 registered attendees… Yes, really unbelievable. It just shows that computer vision is one of those core areas in AI and thanks to technologies like deep learning, the progress on many problems has been amazing over the past few years and we expect to see this trend continuing. IBM is here, of course, because computer vision is so important to a lot of the work that we do in many different industries. We are working on problems in everything from healthcare – think of medical imaging – to retail. We are showing some work here, for example, around using dialog and image retrieval together for applications like fashion and shopping, to even fundamental technologies, like how do we improve the efficiency of inferencing from a chip point of view. You have several papers that are presented here at CVPR; can you tell us about some of them? Yes, sure, we are showing a bunch of things here. One important activity that I will highlight is that IBM has recently built a partnership with MIT – something that we call the MIT-IBM Watson AI Lab. This is a pretty deep partnership on many different areas of AI. It is very collaborative – we don’t necessarily see a distinction between what the people at university do and what IBM researchers do. We are both rolling up our sleeves and working on everything together. One of the first things to come out of the partnership is the development of a unique and very large annotated dataset for actions in video. It’s more than one million videos today and these are annotated at the three-second clip level, with a vocabulary of hundreds of different action categories. This Friday, we will have a workshop here and together with MIT organize the first open evaluation and challenge around this particular dataset. I cannot talk about IBMWatson without asking: what have you got coming up? Certainly, in the area of vision, there is so much more to do. I think we will see a lot more coming out. Just a hint of some directions – we are showing some work here whi ch i s around sports John R. Smith - IBMWatson 4 Guest Thursday