Computer Vision News - May 2024

31 Computer Vision News Coming to terms with the fact that you can do really interesting, impactful work, a lot of times taking work that’s coming out of a research lab, where we’re expanding on that work and bringing it out to the real world, it’s just as good. You don’t have the title. Your parents can’t say, ‘My daughter’s a doctor,’ or something like that, but my parents can say, ‘Look, my daughter worked on all these satellites, my daughter started two companies, my daughter works at a Google subsidiary now.’ These are things that are still on the table if you decide you don’t want to do a PhD. I think there’s a lot of work to still be done about making that kind of education more equitable and accessible to people who don’t come from the most affluent of backgrounds. That’s my quick take on that. Do you think you would have been more of an insider in the computer vision community had you pursued a PhD? Oh, probably, I think, yeah. I guess I’m more motivated by applications than I am motivated by research. It’s more of, ‘Here is this thing that we think we can make in the world, that I think that we can make exist. Is there research that we can leverage to get from where we are to where we want to be?’ That’s a little bit different from, say, a PhD who’s like, ‘Here’s the state of the art. We’re going to push the state of the art just a little bit further over the line. We’re just going to get a little bit higher on this metric.’ That’s good work, and sometimes I think there’s really innovative work, but I’m more satisfied by saying, ‘Hey, we did this thing out in the real world.’ Doing that, I will say, is really difficult with computer vision. If you look at it, there really aren’t that many things that are out there in the world. I mean, there are a lot, but by and large, it’s very difficult to do. How do you see this science evolving in the next decade? I have a slightly contrarian view. I do think that we’ll continue pushing the edge of deep learning, but I suspect, I could be wrong, that some of these deep methods will eventually reach some plateau level. This has been the case with most research since we’ve done research. We have some innovation, that innovation comes to its logical conclusion, and then we Kat Scott