Òòòò Blockin Blockinòò Òòööý Ååòòññþþøøóò Ööññûóöö Óö Ë Blockin Blockinòò Ìüø Êê Blockinóóòòøøóò Ûûøø Ìóô¹¹óûò Ùù×

1 Center for Visual Information Te hnology, IIIT Hyderabad, India. 2 THOTH team, Inria Grenoble Rhone-Alpes, Laboratoire Jean Kuntzmann, CNRS, Univ. Grenoble Alpes, Fran e. *Corresponding author Email: anand.mishra resear h.iiit.a .in Phone: +91-40-6653 1000 Fax: +91-40-6653 1413 Figure 1 A typi al street s ene image taken from Google Street View. It ontains very prominent sign boards with text on the building and its windows. It also ontains obje ts su h as ar, person, tree, and regions su h as road, sky. Many s ene understanding methods re ognize these obje ts and regions in the image su essfully, but overlook the text on the sign board, whi h ontains ri h, useful information. The goal of this work is to address this gap in understanding s enes.