Computer Vision News - September 2020

Anelia Angelova is a Senior Staff Research Scientist and Robot Vision Team Lead at Google. Anelia, what are you doing at ECCV 2020? Quite a few things. This time around, I’m organizing a 4-D vision workshop. That is something very useful for autonomous driving or for robotics. It concerns a 3-D understanding of the scene, in time. That’s why we call it 4-D vision. I’m also giving a talk for an autonomous driving workshop, and the topic of my talk is learning from self and weak supervision. I talk about our recent work that concerns looking into learning with not much sensor supervision, to complement your sensors, or learning with very little labeling, so that you are able to scale your algorithms, especially for self-driving cars. For example, not having to label the segments of individual cars, and people, and so on. And we also have a number of papers accepted. I can broadly speak about them. Please do. The papers cover a number of topics. They are a product of my team, and my team is really amazing. First, we have a paper that continues our architecture search for video understanding. We have started in the last year and half to look a lot at videos, because I believe robots do not obtain information in a static world, and same for autonomous driving. I do believe we really need to look at videos and also take advantage of the long horizon, like what hashappened previously, let’s say 0.1 seconds before my current state, and what has happened one second before, Saints Cyril and Methodius National Library, Sofia, Bulgaria. “We can, using our knowledge and science, pull through and come out stronger!” 2 Women in Computer Vision 6 Best of ECCV 2020