CVPR Daily - Wednesday

DAILY Wednesday Pascal Fua 10 I noticed Mathieu, of course, and I noticed your lab has six papers accepted at CVPR 2020. Another superb achievement! Thank you! Everyone is very productive. It’s nice actually. If you want a bit of personal history: my world, and I think the world of many of my colleagues – especially those of my age – shook in 2012. 2012 is when there was this NIPS and ECCV workshop and deep nets burst into the scene. Maybe I should go a bit further back. When I started my PhD, it was the time when the first wave of the deep nets – they were called neural nets at the time – was essentially dying. What I heard about them, is that they did not work and that I should not touch them. I didn’t and, as result, knew very little about them. Then 2012 happened, and it felt to me as a major tsunami. Yann LeCun and a few others were saying that us classical computer scientists will be swept away by the big wave. We were the dinosaurs. There were a couple of difficult years after that. However, it would appear that the dinosaurs learned how to swim! It rocked the boat, but it didn’t quite sink it. A lot of what we do now is improving on what we used to do. There’s been an obvious jump in performance, but the 50 years of knowledge we have behind us is still useful. I still like geometry and it still has its place. Some of the papers we write have been about epipolar constraints and fundamental matrices. It’s not too late to review how to put them into a deep framework. It can be done and that’s in part what we have been doing. For a dinosaur, you have done very well this year! At a time when the community and numbers of works are growing exponentially, how did you do exponentially, how did you do it? Very easy: Good students! EPFL in Lausanne is such a good school and it attracts some outstanding students. It prides itself on being enormously international. Both at the faculty and the student level. That’s a part of it. We can recruit some of the best people from everywhere. Past PhD student and now scientist and lecturer Mathieu Salzmann How do you recognise great potential in a Master’s student when they first arrive? We use the same filters as everyone else. Good grades from good schools