Computer Vision News - August 2018

future frames. The nice thing about this is that this representation is still highly detailed spatially but it's a higher level than RGB. It's also more useful for the application settings that I was mentioning, and it should be better suited for learning about scene and object dynamics, interactions and things like that. Recently, we extended this method to the prediction of instance segmentation. Basically, to learn a model of object dynamics, we figured it made sense to include in the inputs of our model a notion of object, and semantic segmentation doesn't have that, because it doesn't differentiate between instances. And so this in turn allowed our model to predict much more precise masks. If you're interested, you can check out our ICCV and just accepted ECCV papers online. You can also find our websites, to see examples of our predictions. You told me about testing surprises on babies. Tell me what surprises you! One thing that I have been surprised to see is how fast everything is moving. When I started my PhD, generative adversarial networks had only recently been proposed… now they are all over the place! I don’t think anyone expected how huge and disruptive the 2012 deep learning revolution was going to be. At every corner, it looks like we are on the verge of discovering something important. It’s very exciting! Can you tell us something about Facebook that we don’t know? People don’t always realize how much the lab looks like an academic lab. That’s something that I insist on because that’s pretty special about it. Life as a scientist is challenging for many reasons. Is there one thing that could make it easier? I think I’d like to be able to sleep a lot less than I do. That would save me a lot of time. As a researcher, you always need to learn more about your field. Sometimes it can be pretty terrifying to see the number of things that you don’t know. What drives you to devote so much time in the lab? As a PhD, I guess the aim is that, by the end of your PhD, people can recognize you for the things that you have done. Maybe out of some kind of vanity, we all want to make that kind of contribution to the state-of-the-art and propose something really novel. That requires a lot of work, a lot of reading, a lot of experimentation. That seems like a very brave decision. Did you decide at a young age that you wanted to do this with your life? I was always really into math. I really liked it. I chose my studies for that. Strangely, like quite a few people my age at that time, I thought math was just a selection tool. We didn’t realize it could be useful in real life and real jobs. 26 Women in Computer Vision Computer Vision News Women in Science