Using a Priori Knowledge to Improve Scene Understanding

Brigit Schroeder, Alexandre Alahi; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2019, pp. 0-0

Semantic segmentation algorithms that can robustly segment objects across multiple camera viewpoints is crucialfor assuring navigation and safety in emerging applicationssuch as autonomous driving. Existing algorithms treat eachimage in isolation, but autonomous vehicles often revisit thesame locations. We propose leveraging this a priori knowledge to improve semantic segmentation of images from se-quential driving datasets. We examine several methods tofuse these temporal scene priors, and introduce a prior fusion network that is able to learn how to transfer this information. Our model improves the accuracy of dynamic object classes from 69.1% to 73.3%, and static classes from 88.2% to 89.1%.