DenseCRF_refine_saliency-map

This is a ready-to-use python file for refine saliency maps using dense CRF. If you want to know more about CRF, you can refer to here.

Citation

@article{Saliency-Refinement-DenseCRF,
    Author = {Wei Ji},
    Title = {The application of the DenseCRF for saliency detection.},
    Journal = {https://github.com/jiwei0921/DenseCRF_refine_saliency-map/},
    Year = {2019}
}

Dependencies

environment: You need to run conda create -n SOD python=3.7 .
pydensecrf: This package depends on pydensecrf, available via pip install git+https://github.com/lucasb-eyer/pydensecrf.git.
opencv3: This package depends on cv2, available via pip install opencv-python.
numpy: This package depends on numpy, available via pip install numpy.

Usage

Note : Use Python 3.7

1. put your input (.png) and saliency maps (.png) in images and predictions files, respectively.
2. set your input_path, sal_path and output_path in densecrf_sal.py.
3. run python densecrf_sal.py
4. then, you can see the fine results in the output file.

Maybe Bug

1. pydensecrf/densecrf/include/Eigen/Core:22:10: fatal error: ‘complex’ file not found
   #include
   ^~~~~~~~~
   1 warning and 1 error generated.

Command: conda install -c conda-forge pydensecrf

Related Citations

1. https://github.com/Andrew-Qibin/dss_crf