ECCV 2016 Daily - Friday

4 ECCV Daily : Fr iday Interview with Jitendra Malik Ralph Anzarouth: Jitendra, you have taught computers to do so many things. Is there one thing that computers haven’t learned from you yet and you would like to teach them? Jitendra Malik: In the context of computer vision, I think one of the key problems that we have not yet solved is prediction: given a certain situation, what will happen next. Usually, the benefits of perception are in guiding action. One of the ways in which it comes about is that in any situation we know what is likely to happen next. If we see a tiger moving, we kind of know that it is going to jump at us or whatever… or cars moving. But sometimes, the behavior is more sophisticated. Let’s say you are sitting in a restaurant, and you are waiting at a table. Then you have an expectation that a waiter will come to you. So that’s a prediction. If you are playing some sports, then you have a prediction about whether somebody will pass the ball to somebody else, whether somebody is trying to score a basket and whether that will succeed or not. What I like about the visual prediction problem is that it combines understanding about objects, agents, behavior, short term, long term, and it brings in some amount of cognition, but it is a very visual problem. You must use your own perception of the scene to do this. Ralph: There is also some random component in it. Jitendra: Yes, so it’s not that you will always have the right answer. You have a probability distribution over a set of possible futures. “In the context of computer vision, I think one of the key problems that we have not yet solved is prediction”