Center for Research in Comptuer Vision
Center for Research in Comptuer Vision

Seminar Announcement

Human-Machine Computer Vision Systems Transferring Human Capabilities to Visual Classifiers

Dr. Concetto Spampinato of the University of Catania

Wednesday, October 11, 2017 · 4:00PM · PS160/161

In this talk, I will present several approaches based on different degrees of involvement and awareness, for including humans in automated vision approaches. Per se, this is nothing new: several "interactive" approaches following the "human-in-the-loop" paradigm have been proposed to help solving computer vision tasks - most notably, object segmentation, i.e. extracting object contours from an image. At a more basic level, crowdsourcing has been largely employed in computer vision to annotate image datasets for algorithm benchmarking, by asking users to specify the category (class) or to draw contours of objects in an image. However, these kinds of interactions are still too explicit, by which we mean that users' actions are consciously targeted at the computer vision task they are working at, and as a consequence they need an incentive to motivate them to perform the task, which often amounts to monetary reward. A slightly less explicit user involvement modality is gamification that is the process of putting a job in the form of a game, thus aiming to attract potential participants in a more natural way than crowdsourcing. The difficulty lies in how to make a game out of a job which is inherently non-entertaining, for there would be no need to perform gamification otherwise. In these examples, human feedback is generated through an explicit and voluntary action by an operator working at a task. However, we want to go "deeper" by exploring more implicit modalities for human knowledge and capabilities to back up computer vision, i.e., by integrating a kind of human feedback which comes in the form of either knowledge modelling or physiological measurements taken while performing a task designed to solicit such reactions and to make them useful to automated methods. The underlying idea is that such form of implicit feedback can be potentially more informative than traditional explicit feedback, due to being generated by - and, in a sense, "closer to" - the very underlying physiological processes employed by humans to solve vision tasks, thus switching the paradigm from "human-in-the-loop" to "human-computation systems". In a different but orthogonal sense, we will investigate how the synergy between human potential and automated methods can be shifted from a "top-down" paradigm - where direct user action or human perception principles explicitly guide the software component - to a "bottom-up" paradigm, where instead of trying to copy the way our mind works, we exploit the "by-product" (i.e. some kind of measured feedback) of its workings to extract information on how visual tasks are performed.

Concetto Spampinato received the Laurea (grade 110/110 cum laude) degree in computer engineering and the Ph.D. degree from the University of Catania (jointly with the University of Edinburgh, School of Informatics under the supervision of Prof. R.B. Fisher), in 2004 and 2008, respectively, where he is currently an assistant professor. In 2014 he created and currenlty leads the Pattern Recognition and Computer Vision Laboratory (PeRCeiVe Lab) at the University of Catania. His research interests include mainly computer vision, statistical machine learning and deep learning (mainly for image classification and video object segmentation), medical image analysis, multimedia and vision, with a - recent - particular focus on human-computation systems, which include gamification, visual-knowledge ontology modelling and physiological measurements related to human visual systems. Furthermore, he is one of the pioneers -being one of the ideators of the EU project Fish4Knowledge- of underwater computer vision, i.e., the development of computer vision solutions for monitoring sea and ocean biodiversity. Since October 2017 until February 2017 he was visiting professor at the Center for Research in Computer Vision and the University of Central Florida. He has coauthored over than 100 publications in international refereed journals and conference proceedings. He is member of IEEE, CVF and IARPR.