Computer vision has advanced in leaps and bounds over the past few years, owing largely to the re-popularization of convolutional neural networks. Multiple claims are being heard that machine vision is maturing and will (or already has) surpassed human capabilities. This is far from the truth -- we are only beginning to scratch the surface of truly interesting problems. Very large gaps still exist between human and machine vision - in how they learn and in how well they perform. In this talk, I shall highlight several such gaps (along with approaches to bridge them) in the context of four recent works from my research. I will finish the talk with my thoughts on future research directions to tackle in order to push machine vision towards that of humans.
Amir Rosenfeld received his B.Sc in Computer-Engineering at the Hebrew University of Jerusalem (HUJI) at 2006 and his M.Sc. in Computer Science at HUJI in 2011 Under the supervision of Prof. Daphna Weinshall. He completed his PhD in Computer Science at the Weizmann Institute under the supervision of Prof. Shimon Ullman in 2016. He is currently a joint postdoctoral fellow at the Laboratory for Active and Attentive Vision, York University (hosted by Prof. John K. Tsotsos) and in the University of Toronto (hosted by Prof. Richard Zemel). His main research interests include machine learning, computer and human vision, visual attention and adaptive learning systems.