(CSE Colloquium Lecture Series)
Facebook AI Research
Friday, February 24, 2017
Room 1202, CSE Building
Abstract: To be useful to downstream applications, visual recognition systems must solve a diverse array of tasks: they need to recognize many categories, localize instances of these categories precisely in the visual field, estimate their pose accurately and so on. This set of requirements is also not fixed a priori and can change over time, requiring recognition systems to learn new tasks quickly and with minimal training. In contrast, visual recognition systems today tackle only a few narrowly defined tasks such as object detection or image classification, and need lots of data to learn new tasks. In this talk I will describe my work on removing this shortcoming. I will show how existing recognition systems can be extended to produce richer outputs, such as pixel-precise localization of detected objects. I will also show how we can build recognition systems that can learn new tasks, such as recognizing unseen categories, from very little data.
Bio: Bharath Hariharan is currently a post-doctoral researcher at Facebook AI Research, where he has been working with Ross Girshick, Piotr Dollar, Larry Zitnick and others. Bharath received his PhD in computer vision at the University of California, Berkeley in 2015, advised by Prof. Jitendra Malik.
Related Research Publications
Host: Manmohan Chandraker (firstname.lastname@example.org)