SDM-based probability (SDMP)

Description

An implementation of the "Sampling Distribution of the Mean"-based probability (SDMP). It enables the comparison of MC rendering algorithms (using Mitsuba) as outlined in the corresponding paper.

Getting Started

Dependencies

Python 3:

• numpy
• scipy
• matplotlib
• scikit-image
• scikit-learn
• mitsuba

Installing

1. Setup your Python installation. (for Anaconda users see .conda.yml)

2. Copy or install the this module.

Executing program

1. Setup a Mistuba XML file as the reference, e.g. "./data/veach-ajar/scene-ref.xml".
2. Setup Mistuba XML files for the test / comparison, e.g. "./data/veach-ajar/scene-control.xml" and "./data/veach-ajar/scene-biased.xml"
3. Make sure sampler parameters match for all of the used scenes.
4. Choose SPP (>32 recommneded) and iteration count (e.g. 1024 for the reference and 32 for the test runs) and execute:

python -m sdmp 32 1024 ./data/veach-ajar/scene-ref.xml 256 ./data/veach-ajar/scene-control.xml ./data/veach-ajar/scene-biased.xml

For detailed command line parameters see help:

python -m sdmp -h

Alternatively, one can also run visualize.py to see the renderings and statistics updated during computation.

Authors

Christian Freude, freude (at) cg.tuwien.ac.at

Version History

• 0.9.0
  • Initial Release

License

This project is licensed under the GNU GPL LICENSE - see the LICENSE.md file for details

Acknowledgments

Funded by Austrian Science Fund (FWF): ORD 61