akashbapat/domain_transform_solver

Cuda implementation for domain transform solver, an edge-aware optimization algorithm.

Domain Transform Solver

This code is not the exact version used in the paper, but has undergone many changes.

About

This repository "Domain Transform Solver" is the CUDA implementation for paper "The Domain Transform Solver" published in CVPR 2019. This software is licensed under the BSD 3-Clause license. If you use this project for your research, please cite:

@inproceedings{bapat2019dts,
  title={The Domain Transform Solver},
  author={Bapat, Akash and Frahm, Jan-Michael},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  year={2019},
}

Compilation

You need the following dependencies to run the code.

1. Cuda 8 or 9. Does not work with Cuda 10.
2. OpenCV
   • Mainly used for image IO or in the *cli.cpp, not used in the main optimization code.
3. GFlags
   • Only needed to process the commandline flags. The code is organized in such a way to simply remove the flags at the top of each *_cli.cpp with your favorite cmd flag processor. One of the favorites is Boost.
4. GTest (optional)
   • Turn off tests if GTest is not available. Not so important for using DTS.
5. LibCua (https://github.com/trueprice/libcua)
   • Automatically downloaded.

Installing dependencies

I assume a familiarity with cmake and CMakeLists. The code was tested with CMake 3.5.1

1. Installing OpenCV

   sudo apt-get install libopencv-dev
   

2. Installing Glags:

   git clone https://github.com/gflags/gflags.git
   cd gflags
   mkdir build 
   cd build
   cmake ..
   make -j4
   sudo make install
   

3. Installing Gtest:

   git clone https://github.com/google/googletest.git
   cd googletest
   mkdir build 
   cd build
   cmake ..
   make -j4
   sudo make install
   

Installing DTS

This repository is setup with submodules which need to be also clones using the recursive flag as follows:

git clone --recursive https://github.com/akashbapat/domain_transform_solver.git
cd domain_transform_solver
mkdir build
cd build
cmake ..
make

Running the solver via the commandline interfaces (CLIs)

Once you have built the code you can run the solver examples as follows. When inside the folder domain_transform_solver/build

1. Stereo solver

   ./src/benchmark_stereo_cli --path_to_data_top_dir ../tests/testdata/stereo/
   

2. Colorization solver Change the algorithm_name to "dts" to use domain transform solver.

   ./src/benchmark_colorization_cli --path_to_data_top_dir ../tests/testdata/colorization/ --algorithm_name hfbs
   

3. Depth super resolution solver Change the algorithm_name to "hfbs" to use Mazumdar et al, Hardware friendly bilateral solver.

   ./src/benchmark_depth_super_res_cli --path_to_data_top_dir ../tests/testdata/depth_superres/ --algorithm_name dts
   

Contribution

Feel free to create pull requests if you have a contribution or bugfixes.

Contact

akash@cs.unc.edu