Abstract
Stereo techniques have witnessed tremendous progressover the last decades, yet some aspects of the problem stillremain challenging today. Striking examples are reflect-ing and textureless surfaces which cannot easily be recov-ered using traditional local regularizers. In this paper, wetherefore propose to regularize over larger distances us-ing object-category specific disparity proposals (displets) which we sample using inverse graphics techniques basedon a sparse disparity estimate and a semantic segmentationof the image. The proposed displets encode the fact thatobjects of certain categories are not arbitrarily shaped buttypically exhibit regular structures. We integrate them asnon-local regularizer for the challenging object class ’car’into a superpixel based CRF framework and demonstrate its benefits on the KITTI stereo evaluation. At time of submission, our approach ranks first across all KITTI stereo leaderboards.