Title: My History of Computer Vision: From Early Neural Nets to Boosting to Activity Recognition
Abstract: In the first half of the talk, I will describe the history of computer vision, as I lived through it, in terms of my own research, and the related world events and technological advancements that co-occurred. I will review (and relive!) how the deep neural networks and GPUs of today evolved from the hand-crafted feature detectors and massively parallel vision computers. I will discuss how the early systems were precursors to modern algorithms and how the hardware capabilities and technological advancements played a central role. In the second half of the talk, I will describe my current research in lineage construction for pre-implantation mouse embryogenesis using simulated annealing and a specialized 3D neural. network for mitotic division classification.