Computer Vision News - September 2020

or even at five seconds before. We have a paper on that, that continues with this trend. It’s called AssembleNet. It’s a very, very powerful algorithm for video understanding. We have work to share on placing 3-D objects in the scene from2-D images. It’s called Mask2CAD. This is very exciting work as well, because it is helping you create almost like a virtual world from what you really see. That can be very useful for robotics as well. We have work on unsupervised flow for which we already have an open source algorithm that is state-of-the-art, very simple and fast as well. Wealsohavework toshareonprediction of future activity and human pose prediction that can go much further in the future than previous approaches. It’s using a treatment of the stream, either video stream or pose stream, modeling the consecutive events as a grammar. For example, from a state, I can go to three other states, which I can model by a rule. How many people are in your team? About twelve: an amazing team! We cover a very broad set of topics. The last paper was also an architecture search, but this time focusing more on attention and that is a joint work with our collaborators in Google Cloud. Did you hire all the twelve people? No, not exactly. I initially inherited a team of 6, and I have grown the team. You’re the boss? Yes, I manage the team, and I'm what they call at Google ‘Tech LeadManager’. What kind of manager are you? [ laughs ] Well, that's an interesting question. You should ask my team! [ both laugh ] I would say that I am too democratic, and that actually fits the team well. We’re mostly research scientists or research engineers. With that, comes some freedom of what topics to choose and how to approach them. The result is that we can cover very broad topics within computer vision and robotics. Is this very typical of Google, or is it special to your team? That's a really good question. It's very typical of Google to allow research freedom for our researchers. Google is a very bottom-up company. No matter if you are in the research organization “… it is a truly special place. What makes it special are the people and their spirit.” 153 Anelia Angelova 63 Best of ECCV 2020

RkJQdWJsaXNoZXIy NTc3NzU=