Computer Vision News - December 2018

35 Computer Vision News Lydia Chilton Women in Science and I don’t have any tools from the beginning. I just want to spend a bunch of time figuring it out. I like the pleasure of finding things out. I found that I have the freedom to do that in the academia. It is very possible that if I looked harder, had the right connections, or just had a better first experience or a different first experience in the industry, I would have found that as well. You don’t have time to search forever. If you find something that works, it’s good enough. I enjoy teaching as well! I enjoy undergraduates and working with them. I use them as a great way to tell whether people think the research is cool or not. What is the dreamoutput of your work? There are two possible things that would be nice. One is a fully automated system where you input the message that you want to convey and it gives you an image. Here is a visual metaphor that we have already tested. It has “such and such” a click-through rate, and it’s done. That would be impressive if we could totally understand those problems to the degree that it can be done without any input. That is the sort of traditional artificial intelligence, machine learning, and computer vision way of seeing things. I think that communication, particularly with humans, is very hard to fully decompose. Even all the images on Google don’t take into account all the things that we see in everyday life. They are very biased towards things that are exciting enough to take a picture of. Another fantastic outcome for me would be a decomposition of the problem. We get the computer vision to do some of the parts that are hard for people like mocking up the results, creating that final image, doing a big search for symbols like what represents fire, and hot, and energy, and all these things. Computers can help with those. Humans continue to drive the search. We’d like to have more feedback in the system, decomposing the problem and making it far easier for a person to do. A person could spend like 30 seconds guiding the search and then redirecting it when its wrong. Tell us about China, where you lived three times. There are some parts about China that I really enjoyed. Excellent food, obviously. Microsoft Research Asia was amazing. It has a very international group. Obviously, there are a lot of Chinese people there, but one of my bosses was from England. There was a designer from Italy, and a graphic artist from New Zealand. There were a lot of people who ended up there. I always felt welcome. I speak Chinese. I don’t read or write. That’s the hard part, but I can order food. I got a lot out of the multicultural nature of it. I met a ton of friends there. A lot of interesting people got together. It was where I first got exposed to design. They were mostly working on physical objects and augmented and virtual reality application, how you can use technology as a means of communication. I never saw it that way before, but that is very important. This CVPR workshop was the first one. Where do you expect this work to go in the future? This is a very applied problem. I hope more advertising companies will get “I like the pleasure of finding things out!”