cvlab-epfl / densecrf

a fork of the densecrf package implementing alternative inference scheme

Description

This is a modified version of a forked densecrf, which was used as a part of the DeepLab.

For more details about the inference algorithm used in this version, please refer to and consider citing the following paper:

@article{baque2015principled,
  title={Principled Parallel Mean-Field Inference for Discrete Random Fields},
  author={Baqu{\'e}, Pierre and Bagautdinov, Timur and Fleuret, Fran{\c{c}}ois and Fua, Pascal},
  journal={arXiv preprint arXiv:1511.06103},
  year={2015}
}

If you are using densecrf, please consider citing the following paper:

@inproceedings{KrahenbuhlK11,
  title={Efficient Inference in Fully Connected CRFs with Gaussian Edge Potentials},
  author={Philipp Kr{\"{a}}henb{\"{u}}hl and Vladlen Koltun},
  booktitle={NIPS},      
  year={2011}
}

If you are using DeepLab, please consider citing following paper:

@article{papandreou15weak,
  title={Weakly- and Semi-Supervised Learning of a DCNN for Semantic Image Segmentation},
  author={George Papandreou and Liang-Chieh Chen and Kevin Murphy and Alan L Yuille},
  journal={arxiv:1502.02734},
  year={2015}
}

Building and Dependencies

You should have matio library installed.

To build the binary, just run make.

Usage

... to be filled in ...

For the complete pipeline for semantic segmentation, please refer to DeepLab.

For the details (parameters) specific to this version, refer to refine_pascal_nat/dense_inference.cpp.