SceneCut: Joint Geometric and Object Segmentation for Indoor Scenes

Update:

• Add code to generate region hierarchies.

Usage

• Run demo.m

Generate region hierarchies from images/ucms

• Download and install COB network https://github.com/kmaninis/COB, place them at includes/COB. Make sure COB path is added.
• Pre-computed ucms for the NYU dataset can be downloaded from http://www.vision.ee.ethz.ch/~cvlsegmentation/cob/code.html. Otherwise run img2ucms function (in COB) to generate ucms.
• Run ucm2tree to generate region hierarchies and node features.

Citation:

If you use this code, please consider citing the following papers:

@Inproceedings{Pham2017,
   author = {{Pham}, T. and {Do}, T.-T. and {S{\"u}nderhauf}, N. and {Reid}, I.
    },
    title = "{SceneCut: Joint Geometric and Object Segmentation for Indoor Scenes}",
    booktitle={2018 IEEE International Conference on Robotics and Automation (ICRA)}, 
    year={2018}, 
}

If you encounter any problems with the code, please contact the first author. Enjoy!