Wiry Nigel, Lucy Fe, Nvidia Research discussed on The AI Podcast
Do. And so i was just going to ask this kind of forgive me for my own understanding of it but is that why it's easier from video because you have so much data you know somebody frames that you can extrapolate data points from two to create those multiple perspectives. Right wait for wide. Wiry nigel correspondence. Because you have many frames and it's easy to track the same point using many techniques like your floor or something else but actually we also need image level correspondence which means for different human or different bird. We also want to know what his correspondence and then our model can generalize to category. And that's actually more easy. Oh more generalize to facilitate a large variety off animals vicious to be reconstructed for cool. We're speaking today with seafarer. Lucy fe is a senior research scientist at nvidia research. And we're talking about a project that she's been involved with for a while now called online adaptation for consistent mesh reconstruction in the wild Taking two d images and video streams and creating three d representations three meshes from them and so in the wild. You mentioned it's Animals birds exclusively that. You're working with. Yes so a burden is actually very typical animal. That people like to work on. So the special thing about bird is that he can always easily deformed for mafia but but actually in our work also extended to some adam. Animals lexi bri. If you check the national around yeah yeah. So i will go. is we. Hope to like empower each and applied to more endangered species which should be very like help for for those kind of researches. Iran in like researching endangered vicious dotted in terms of creating representations of how these species behave in the wild or just being able to spot them. Or what's the application with endangered species so for example for sweetie. The matchel is very hard to even oakton. They're like sweetie shape or three d shaped template from those fishes as that will. The reconstruction will at least the benefit to help the researchers in that area in learning the endangered species to build up a physical model and at live that will help to establish their research On cop out that got it. Let's shift gears for a second and talk about your background if we could. You mentioned you know having a kind of a standing interest in unsupervised learning But tobacco up a little bit more. How did you get into working with a and deep learning and even sort of computer science and the broader field to begin with. Okay yeah and from china and actually go to university like sixteen years ago. My orange major is automation. And at that time i think is not that popular and the for the nation. People usually study the control signs that or like digital media processing face recognition. Those kind of directions. Still one branch in the automation. So usually in the year of our like undergraduate we are allowed to explore our preferred branch off contra science nation while i actually find myself particularly interesting digital image processing you know like goes like preferred to aditya theses images also faints and at that time those kind of tours are not so powerful so basically from there was trying to find internships and i was lucky enough to be immediate by the cassia. Which is the top institute in china. In back i and also like digital image processing or all kinds of things so that i think that's the start. I begin to do the research. Ns actually fourteen years ago. Okay and so How and when did you wind up joining nvidia. So it was four years ago. I was i a intern student. There and then. I joined as a full-time..