Efficient Facade Segmentation Using Auto-context

This package contains the code for "Efficient Facade Segmentation Using Auto-context". If you use this code, please consider citing the following papers.

  title={Efficient facade segmentation using auto-context},
  author={Jampani, Varun and Gadde, Raghudeep and Gehler, Peter V},
  booktitle={2015 IEEE Winter Conference on Applications of Computer Vision},

  title={Efficient 2D and 3D Facade Segmentation using Auto-Context},
  author={Gadde, Raghudeep and Jampani, Varun and Marlet, Renaud and Gehler, Peter V},
  journal={arXiv preprint arXiv:1606.06437},

The code has been tested to work on Ubuntu 14.04 using Matlab R2012a. Please follow the below steps to use this code. 1) run This will install the Darwin library and the Piotr's Computer Vision Matlab Toolbox. Note that the Graz dataset is also downloaded. 2) run src/ This will use the auto-context framework to train and test on the Graz dataset.

  • Upon running the code successfully, you should get around on the Graz dataset.
  • The training may take several hours. If you have more multiple-cores on your machine consider changing the "drwnThreadPool" attribute. Currently it is set to 8.
  • Results are stored in out/graz/fold[1-5]. The dataset is divided into 5 folds (see data/graz folder). For fold1, set[1,2,3,4] are used for training and set5 for testing. For fold2, set[1,2,3,5] as training and set4 for testing.