5 Computer Vision News of robustness based on the loss distribution in the previous steps, ” Sara explains. “ We also incorporated the spatial consistency of the loss by filtering and using diffusion with kernels of different sizes and patches during the optimization. ” Ultimately, RobustNeRF represents a straightforward modification to the loss function in NeRF models. It only requires a few lines of code, so by simply substituting the reconstruction loss with the robust loss proposed here, any NeRF model can be adapted. Sara was one of the authors of a paper we featured in Computer Vision News recently called nerf2nerf by Lily Goli , as part of Andrea Tagliasacchi’s team at the University of Toronto. She has also worked closely with leading deep learning figure Geoffrey Hinton at the University of Toronto and Google. Can she give us an insight into the man referred to as one of the ‘ Godfathers of AI ’? “ When we started the team in Toronto, Geoff and I used to have lunches every day together , ” she recalls. “ He told me a story about when he was working in a woodworking workshop. At first, he didn’t feel like going to university. He said he learned a lot in the year or two he spent there. He then wanted to learn more about science, so he started in an area related to chemistry or physics, but after a couple of years, realized he was not interested in that and changed to cognitive science. He had a very explorative approach to getting to where he is now! ” RobustNeRF: Ignoring Distractors with Robust Losses
RkJQdWJsaXNoZXIy NTc3NzU=