I am presently working as an Assistant Professor at the Department of Electronics & Electrical Engineering, IIT Guwahati. Previously I was working as the Team Leader of the Computer Vision Group at TCS Innovation Labs (Delhi). I have obtained B.E. in Electrical Engineering from Jadavpur University, followed by M.Tech. & PhD. from IIT Kanpur (Electrical Engineering). My research interests lie in the area of Computer Vision and Machine Learning.
If one takes the grand goal of computer vision as that of enabling an agent to respond appropriately to the widest possible range of situations then it becomes clear that in the end, no system can be engineered to handle the vast diversity of inputs that may arise. We explore the possibility that in future, vision systems may not be so much “programmed” directly by the engineers using code but might be “meta-programmed” so that they may discover the salient aspects of different visual contexts through repeated exposure, along with (minimal) inputs from supervisors or peers. In this view, the vision system is initially an observer, accumulating and identifying similarities between visual sequences leading to a set of incipient visual categories of objects, actions and relations. Occasionally, it may “reify” or consolidate some of these categories based on interactions with other learning agents where they may refer directly to some of the objects/actions, or may only discuss these in a broader context, as in language/speech based descriptions of the scene. Based on these limited exposures, the vision system must form its categories as best as it can.