Section: New Results

Region-based Semantic Segmentation

Paticipants: Pawan Kumar

In  [9] we consider the problem of parameter estimation and energy minimization for a region-based semantic segmentation model. The main problem we face in the context of energy minimization, is the large number of putative pixel-to-region assignments. We address this problem by designing an accurate linear programming based approach for selecting the best set of regions from a large dictionary, which is constructed by merging and intersecting segments obtained from multiple bottom-up over-segmentations. The lack of fully supervised data is tackled by using a latent structural SVM formulation, where the latent variables model any missing information in the human annotation. Using large, publicly available datasets we show that our methods are able to significantly improve the accuracy of the region-based model.