Gradient Outlier Removal for Gradient-Domain Path Tracing

Saerom Ha, Sojin Oh, Jonghee Back, Sung-Eui Yoon, Bochang Moon

Overview

This code is the implementation of the method demonstrated in the paper Gradient Outlier Removal for Gradient-Domain Path Tracing by Ha et al. It is based on gradient-domain path tracing (G-PT) by Kettunen et al., and improves the rendering quality by adding gradient outlier detection & removal process to G-PT. For more details, please refer to our project page.

Our solution is separated into two parts.

dependencies implemented on top of G-PT
our gradient outlier removal project (RobustFit)

The first one is implemented on top of G-PT to get sample buffers, so G-PT framework which extended Mitsuba 0.5.0 should be set up beforehand. Please refer to the G-PT framework (https://github.com/mmanzi/gradientdomain-mitsuba.git) for more details.

Using the sample buffers, our method (the second one) conduct reconstruction with detecting and rejecting gradient outliers. This code is implemented and tested using Visual C++ 2013 and CUDA Toolkit 9.0 in Windows 10. Unfortunately, Linux and Mac OS are not supported yet.

If there is any problem, question or comment, feel free to contact us:
Saerom Ha(saeromm.ha@gmail.com), Sojin Oh(jaisonoh@gist.ac.kr) or Jonghee Back(jongheeback@gist.ac.kr)

Usage

Build and Run

Gradient-domain rendering framework must be prepared first.

Prepare gradient-domain path tracing (G-PT) framework
Build RobustFit project and copy RobustFit.lib and cudart_static.lib into dependencies folder of G-PT
- Copy into $(GPT_PATH)/dependencies/lib/x64_vc12/
- cudart_static.lib is located on $(CUDA_PATH)/lib/x64/
Copy source code files in the RobustFit project (9 code files) into dependencies folder of G-PT
- Copy into $(GPT_PATH)/dependencies/include/RobustFit/
Unzip robustfit_dependencies.zip and overwrite the files in G-PT framework
Build gradient-domain path tracing (G-PT) and run mitsuba.exe the same as the usage of G-PT

Scene

Please make sure the xml scene files include attributes for using G-PT, as the scenes can be downloaded from G-PT project homepage (https://mediatech.aalto.fi/publications/graphics/GPT/).

License

All source code files in robustfit_dependencies.zip are released under the GNU GPLv3. A software list that we use is as follow.

Mitsuba 0.5.0 released by Wenzel Jakob under the terms of the GNU GPLv3

Separately, the RobustFit project is under a BSD License. Please refer to our license file.

Citation

If you use our code or paper, please check below.

@article{Ha19,
  author = {Ha, Saerom and Oh, Sojin and Back, Jonghee and Yoon, Sung-Eui and Moon, Bochang},
  title = {Gradient Outlier Removal for Gradient-Domain Path Tracing},
  journal = {Computer Graphics Forum},
  year = {2019}
}

Release Notes

v1.0

Initial version of robust-fit.

CGLab-GIST/robust-fit