Speaker: Ravi Kiran Sarvadevabhatla

Date/Time/Venue: Feb 19, 12noon, SIT001

Title: Looking at (Almost) Nothing, Understanding Everything: Deep Learning for Hand-drawn Sketches

Abstract: Deep Learning-based object category understanding is an important and active area of research in Computer Vision. In the talk, we shall see deep-learning approaches for analysis of hand-drawn object sketches and modelling sketch-driven cognitive processes. On the analysis front, I propose a hierarchical approach for the challenging problem of sketch parsing (i.e. given an object sketch, determine fine-grained details -- pose, semantic parts etc.). As an example of modelling sketch-driven cognitive processes, I shall introduce the first computational model for Pictionary, the popular word-guessing social game. Towards the end of the talk, I shall also briefly overview deep-learning based approaches for other sketch-centric problems and discuss two useful perspectives for studying sketches in general.

Speaker Bio: Ravi Kiran Sarvadevabhatla is a Research Associate at Qualcomm India. He recently submitted his Ph.D. thesis from Department of Computational and Data Sciences, Indian Institute of Science (IISc), Bangalore and was advised by Dr. R. Venkatesh Babu. Before joining IISc in 2014, he worked as a Research Engineer at Duos Technologies, Jacksonville, USA (2012-2013). Prior to that, he was a Research Engineer at Honda Research Institute USA, Mountain View, USA (2008-2012). He obtained his MS in Computer Science from University of Washington, Seattle, USA in 2008. He got an integrated MS/B.Tech degree (Honors) in Computer Science from International Institute of Information Technology, Hyderabad (IIIT-H) in 2004. His research interests include visual object category understanding, document image analysis, multi-modal human-robot and human-computer interaction.  For additional info, visit his webpage at https://ravika.github.io/.