Urban scene segmentation by graphical model

This paper propose a simple and flexible frame work, using graphical model to understand diversity of urban scenes with varying viewpoints. Our algorithm constructs a CRF network using over segmented superpixel regions and learn the appearance model from different set of features for specific class of our interest. Also, we introduce a training algorithm to learn a model for edge potential among these superpixel areas based on their feature difference. The algorithm gives competitive and visually pleasing results for urban scene segmentation. We show the inference from our trained network improve the class labeling performance compared to the result when using the appearance model solely.