CVPR Daily - Tuesday

Federico Pernici is an assistant professor and researcher at the University of Florence. He talks to us before his poster presentation today. Federico describes his work as a lifelong learner in which we are learning the multi-object appearance of faces . It’s multi-target tracking, where we are tracking multiple faces, but able to learn their appearance in a lifelong way. The idea is once they exit the field of view and re-enter a few minutes later, we can provide the same ID. He says the challenging part is that video does not provide IID data. Typically, in supervised learning, you have IID data. They are already labelled and are generated from an unknown distribution but in an IID. Humans have already selected the best patterns. In video, you are always viewing the same person with the same appearance, so after a few seconds, you don’t need any more data on that person. You have to throw data away instead of collecting it, because it is redundant . The basic idea is to reduce redundant data. You have to throw data away instead of collecting it, because it is redundant. The basic idea is to reduce redundant data. 8 Federico Pernici Tuesday