CVPR Daily - Tuesday

you are going to do. Can you tell us about your future plans? During my PhD, most of my research was on 3D perception, trying to understand the 3D structure of the environment. For example, object detection is not just outputting a 2D bounding box around object but also giving them the 3D dimensions and physical location. Through that process, I started to realize a very good application of why we need a 3D understanding of robotics. For robots, they need to move in the physical world. They need to interact with 3D objects. That's a very good motivation as to why we need 3D. That’s the motivation behind why I got into robotics! The more I was working in robotics, I realized there are many different ways that we can reformulate computer vision power. That’s very interesting. This gave me a new perspective as to how I think we should approach computer vision power. What is the most exciting thing in computer vision and the most exciting thing in robotics? Lately, I am mostly interested in how we can use robots' active actions, like the robots actually pushing objects, picking up objects, and interacting with objects, using those active interactions to help improve computer vision power. For example, if we want to detect objects, if they are kind of cluttered, like stacked in a pile in a bin, it’s very hard to recognize them. But, if you have a robot pick them up and isolate them from the clutter then it will become a lot easier to recognize them and process the information. It’s kind of like using the robot's interaction to help us to make the computer vision power easier. I’m sure that this award has given you a lot of motivation and satisfaction. Why do you think they chose you? Computer vision with robotics in the past 20 years has been doing their thing in isolation. So in robotics, people always assume that computer vision is just going to solve their problems and give them all their predictions. Then they can add on top of that. But, there is now a lot of work that can go in between. 23 DAILY CVPR Tuesday Shuran Song

RkJQdWJsaXNoZXIy NTc3NzU=