ECCV 2022 - Wednesday

indirectly, it’s beneficial for you too. That’s something I implement as much as I can. I try to make everyone in my team successful. One work I think is really cool is where we fit models to observations – not necessarily of humans but meshes. It’s called Learned Vertex Descent. Instead of doing it the classical way, where you have an objective function, take gradients in parameter space, and then find local minima, we go much more direct. For every vertex in the model, we learn to displace it towards the ground truth iteratively. It’s a super simple idea that leads to good results. Another paper I find cool is about posed neural distance fields: Pose- NDF. Neural fields have been used for representing 2D surfaces embedded in 3D. In this paper, we thought that was limiting. You could generalize and consider those neural fields to represent more general manifolds in higher dimensions. We focus on modeling the manifold of poses. The manifold is like a hypersurface, where every real pose lives in this hypersurface. As soon as you leave this hypersurface, it’s the wrong pose. We use this idea to learn the manifold using neural fields so that, for example, we can take a completely wrong pose and then project it back to the manifold with this machinery. [ Just after this interview, this paper won the Best Paper Honorable Mention at ECCV 2022. Kudos! ] What can you tell us that is special about the work of your lab? I want the lab to have a collaborative spirit so that everyone has their own project, but we have common long-term goals. That gives us some leverage because then people can help each other and reuse code for different projects. This collaborative spirit is a strength that I cherish very much. How did an authentic Catalan like you become a prominent computer science scientist in Germany? What is the meaning of this? 12 Gerard Pons-Moll Your group has an impressive number of papers at ECCV this year. Can you tell us about one of which you are particularly proud? I don’t like to choose because I think they’re all excellent, but if I have to, can I pick two? Of course!