Efficient Structured Parsing of Facades Using Dynamic Programming

IEEE Int. Conf. on Computer Vision and Pattern Recognition 2014

Abstract

We propose a sequential optimization technique for segmenting a rectified image of a facade into semantic categories. Our method retrieves a parsing which respects common architectural constraints and also returns a certificate for global optimality. Contrasting the suggested method, the considered facade labeling problem is typically tackled as a classification task or as grammar parsing. Both approaches are not capable of fully exploiting the regularity of the problem. Therefore, our technique very significantly improves the accuracy compared to the state-of-the-art while being an order of magnitude faster. In addition, in 85% of the test images we obtain a certificate for optimality.

Publication

Results

ECP dataset

The original dataset can be found on http://vision.mas.ecp.fr/Personnel/teboul/data.php. We use the annotations provided by http://homes.esat.kuleuven.be/~amartino/datasets/ECP_newAnnotations.zip.

eTrims dataset

The original dataset can be found on http://www.ipb.uni-bonn.de/projects/etrims_db/.

Miscellaneous dataset

This dataset is composed by different images from facades from all around the world. Most of them can be found at http://vision.mas.ecp.fr/Personnel/teboul/data.php. A compilation of all this images can be downloaded [here]. Another part of this dataset consists of a selection of rectified images from the ZuBud dataset. These rectified images can be found [here].