Section: Software and Platforms

Object category localization

Participants : Ramazan Cinbis, Matthijs Douze, Cordelia Schmid, Jakob Verbeek.

We developed an object category localization system based on a Fisher vector representation over densly extracted local SIFT descriptors [18] . To improve the robustness with respect to background clutter in the detection windows we developed an approximate object segmentation method that is used to weigh the contribution of local SIFT descriptors. Our system achieves state-of-the-art localization performance as measured on the PASCAL VOC 2007 and 2010 datasets. The system is developed in both C, python, and Matlab. The system will be released in early 2014.