Computer Vision News - October 2018

Michelle describes the next steps for this work: they include extending the framework to learn temporal connections between nodes across video frames. Finally, we ask Michelle what it is like to work with Fei-Fei Li : “ Being able to work with Fei-Fei is honestly a really great honour, because she is so knowledgeable in computer vision and has so many years of experience. Learning with her and working with her, I’m able to always work on what we define as the most impactful problems in computer vision. We believe that with all our work, we want to be able to drastically shape the field of computer vision. ” Daily Tuesday Michelle Guo 17 One of the biggest challenges in all of this, Michelle points out, is finding the proper datasets to work with. To her knowledge, it is the first time that this task is being defined at all, so there are very few datasets that allow extraction of the 3D structure . Particularly because they are focusing on human object interactions and relationships in 3D videos. There are very few datasets that are able to provide the labels for this. She adds that this makes the problem perfect! Their task is fewshot learning, so the fact that they have to deal with scarce data makes it a challenging problem for them to address. “A really great honour!”