Computer Vision News - November 2020

240 from it because it's not fine grained enough. It’s also very expensive. Data is expensive, and thenalsometrics. Coming up with good metrics is very complicated as it's much easier to say, “Is there a cat in this image or not?” Whether it is to say, “What is the metric that I use when I'm looking at you so I know you are interested in the conversation?” What exactly are the muscles in your face doing, or your posture? Basically, those three things are the things we need to figure out in order to make progress. Which one of these three things are you the most confident to figure out? I'm most confident about data because there's a lot of visual data that is captured in the world. You just need to look in the right direction and collect the right data to at least start making progress in these directions, to look for things that are not the easiest things to find but the things that would actually answer the questions that you have. For example, now we're looking at conversations that people have on online platforms, to capture a conversation between two people. You can look at Good Morning America videos, where you have two anchors talking to each other. The data exists for things like this. You just have to collect it. That’swhy I was more confident with that. Annotation, I think, is going to be impossible in the general cases. This idea that we can learn everything in a supervised way is just not going to work here. It's not even that we can't pay people enough to do this kind of meditation for us, which we Women in Computer Vision One of the famous hikes wit h Alyosha Efros. With the gray jacket is Angjoo Kanazawa “… it always works out at the end!”