SBD Dataset proprocessing code for CASENet

The preprocessing code is released under the MIT License (refer to the LICENSE file for details).

The repository contains the preprocessing code of the SBD dataset for CASENet. CASENet is a recently proposed deep network with state of the art performance on category-aware semantic edge detection. For more information about CASENet, please refer to the arXiv paper and the paper published in CVPR 2017.

If you find CASENet useful in your research, please consider to cite:

@inproceedings{yu2017casenet,
    author = {Yu, Zhiding and Feng, Chen and Liu, Ming-Yu and Ramalingam, Srikumar},
    title = {CASENet: Deep Category-Aware Semantic Edge Detection},
    booktitle = {Proceedings of the IEEE conference on computer vision and pattern recognition},
    Year = {2017}
}

@inproceedings{yu2018seal,
    author = {Yu, Zhiding and Liu, Weiyang and Zou, Yang and Feng, Chen and Ramalingam, Srikumar and Kumar, B. V. K. Vijaya and Kautz, Jan},
    title = {Simultaneous Edge Alignment and Learning},
    booktitle = {Proceedings of the European Conference on Computer Vision},
    Year = {2018}
}

Note: In this part, we assume you are in the directory $SBD_PREPROCESS_ROOT/.

1. Download the SBD dataset tarball to data_orig/ and untar the dataset subfolder.

   wget -O ./data_orig/benchmark.tgz "http://www.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/semantic_contours/benchmark.tgz"
   tar -xvzf ./data_orig/benchmark.tgz -C ./data_orig/ benchmark_RELEASE/dataset --strip 2 && rm ./data_orig/benchmark.tgz

2. Run the matlab code to preprocess the data.

   # In Matlab Command Window
   run code/demo_preproc.m

   This will perform data augmentation, and generate the .bin edge labels and the corresponding file lists that could be read by CASENet in data_proc/.

The repository of the Cityscapes preprocessing code can be found here.