Computer Vision News - December 2022

16 Humans in Videos Some happened organically, some I chose, and some are a continuation from my postdoc. Human motion synthesis is one that I determined. Is this a passion of yours? Yes, I found a space which is really unexplored. Of course, once you start exploring it, other people do too, so it’s not as underexplored anymore. The text- to-motion synthesis was something I came up with as I felt it hadn’t been done before. That’s where you type some text and want to generate a 3D avatar that does the instruction, like raising your arm, for example. For the benefit of a non-scientist like me, why does the real world need something like this? Gül, what is your work about? Computer vision, first of all, but mainly videos, and particularly, humans in videos. This could be sign language applications, human motion synthesis, action recognition from video, text-to- video retrieval, and a lot of things that nowadays contain vision and language. Do you consider that you are doing something very specific, or are you covering a wide area? I’m covering multiple specific things, which becomes a wide area in the end! [ she laughs ] Did you choose them? Gül Varol is a Research Faculty at École des Ponts ParisTech.