ECCV 2016 Daily - Thursday

predict and so I will work on things that I think are interesting, problems that are unsolved, and then see what happens to them. You let them go and then the world figures out how to use them. I was able to put my code on an FTP site and people could download it. That was already a step that, when I started my PhD, would have been harder. Ralph: If you didn’t work on optical flow, what would you have dedicated your last 30 years to? Michael: I work on several other things other than optical flow. When I graduated, I thought, I can’t just keep working on optical flow. That’s what my advisor did, and I needed to do something a little different. What interests me is how things move. I’m interested in human motion and animal motion. I started working on the motion of the human body. I started with faces, facial motion, recognizing expressions, and then moved on to articulated motion and continuing that today. I look at the motion of people, soft tissue motion, their muscles, and their clothing. Now, we’re extending that to work on animals of all kinds. I think that there is an endless variety of ways nature and creatures in the world have figured out how to move. I’m interested in capturing that and, hopefully, we’ll use it to study animal movement and human movement. At the moment, I am slightly more interested in 3d shape and how to represent it. I think we have a lot of representations for images. We have a very simple representation in 2d grid. A lot of our processing is in 2d representations, but the world is 3d. I’m interested in representations of that 3d world that make processing on it easy. I think that many of the problems we have about reasoning and motion in 2d images is because we’re reasoning about motion in 2d images, and not reasoning about motion in a 3-dimensional world. If we have the right representations of the 3d world, it would simplify many of the problems. Ralph: Is the problem in our reasoning or in the tools that we are using? Michael: In the tools. Ralph: So we can solve it? Michael: Absolutely, and so what is difficult now in terms of representing consistency across time and correspondence across time may be much simpler if we reformulate the problem away from the 2d image grid to something about the 3d world. That means solving a lot of problems. 2d images are very easy to deal with; and some 3d representations in the world seem much more complex: maybe it's meshes, maybe it's something else. But sometimes making it more complex simplifies it in other ways. Ralph: Will we ever master time? Michael: I can’t even master my schedule! I have an assistant that helps me schedule my meetings. I can’t even schedule my day, so no. It’s hopeless for me… [laughs] Interview with Michael Black 7 ECCV Daily: Thursday