This webpage contains datasets and supplementary information for the following paper:
B. Fröhlich, E. Rodner, and J. Denzler. A Fast Approach for Pixelwise Labeling of Facade Images. International Conference on Pattern Recognition (ICPR). 2010
Due to the small number of images available in the eTRIMS database, we generated a similar database using LabelMe which contains a huge number of images with labeled polygons. Since this is a subset of LabelMe images, the images were originally collected by the authors of this publication . All images should only be used for non-commercial and research experiments. Please check with the authors of the LabelMe dataset, in case you are unsure about the respective copyrights and how they apply.
Three ground-truth examples from the dataset featuring label information for classes sky, building, road, pavement, vegetation, window, door and car for the task of semantic image segmentation. Black areas belong to an additional background class.
For our dataset we extracted images which contain buildings, windows, sky and a limited number of unlabeled regions (maximally 20% covering of the image). This procedure resulted in 945 images. The pixelwise labeled images are created by utilizing the eTRIMS categories and a simple depth order heuristic. Due to some label ambiguities the following class names are used as synonyms for corresponding eTRIMS categories: house (building); sideway, sidewalk (pavement); street (road); tree (vegetation). We split this dataset into 100 images for training and 845 images for testing.