MVS-Texturing

Welcome to our project that textures 3D reconstructions from images. This project focuses on 3D reconstructions generated using structure from motion and multi-view stereo techniques, however, it is not limited to this setting.

The algorithm was published in Sept. 2014 on the European Conference on Computer Vision. Please refer to our project website (http://www.gcc.tu-darmstadt.de/home/proj/texrecon/) for the paper and further information.

Please be aware that while the interface of the texrecon application is relatively stable the interface of the tex library is currently subject to frequent changes.

Dependencies

The code and the build system have the following prerequisites:

• cmake (>= 3.1)
• git
• make
• gcc (>= 4.6.3) or a compatible compiler
• libpng, libjpg, libtiff

Furthermore the build system automatically downloads and compiles the following dependencies (so there is nothing you need to do here):

• rayint https://github.com/nmoehrle/rayint
• Eigen http://eigen.tuxfamily.org/
• Multi-View Environment http://www.gcc.tu-darmstadt.de/home/proj/mve/

The following is only downloaded if you use this software for research purposes and thus provide the -DRESEARCH=ON flag (see compilation section below).

• multi-label graph cut optimization http://vision.csd.uwo.ca/code/

Compilation

1. git clone https://github.com/nmoehrle/mvs-texturing.git
2. cd mvs-texturing
3. mkdir build && cd build
4. Generate make file
   • cmake ..
   • IMPORTANT: For research purposes only you can use cmake -DRESEARCH=ON .. instead. This downloads and links against Olga Veksler et al.'s multi-label graph cut optimization, which tends to find better optima and gives better texturing results. However, it is patented and can only be licensed for non-research purposes by the respective authors. For non-research purposes you have to stick to not using the RESEARCH flag. This will use Loopy Belief Propagation instead of Graph Cut Optimization. Also see the license section below for details.
5. make (or make -j for parallel compilation)

If something goes wrong during compilation you should check the output of the cmake step. CMake checks all dependencies and reports if anything is missing.

If you think that there is some problem with the build process on our side please tell us.

If you are trying to compile this under windows (which should be possible but we haven't checked it) and you feel like we should make minor fixes to support this better, you can also tell us.

Execution

As input our algorithm requires a triangulated 3D model and images that are registered against this model. One way to obtain this is to:

• import images, infer camera parameters and reconstruct depth maps using the [Multi-View Environment] (http://www.gcc.tu-darmstadt.de/home/proj/mve/), and
• fuse these depth maps into a combined 3D model using the [Floating Scale Surface Reconstruction] (http://www.gcc.tu-darmstadt.de/home/proj/fssr/) algorithm.

A quick guide on how to use these applications can be found on our project website.

By starting the application without any parameters and you will get a description of the expected file formats and optional parameters.

Troubleshooting

When you encounter errors or unexpected behavior please make sure to switch the build type to debug e.g. cmake -DCMAKE_BUILD_TYPE=DEBUG .., recompile and rerun the application. Because of the computational complexity the default build type is RELWITHDEBINFO which enables optimization but also ignores assertions. However, these assertions could give valuable insight in failure cases.

License, Patents and Citing

Our software is licensed under the BSD 3-Clause license, for more details see the LICENSE.txt file.

IMPORTANT: Using the -DRESEARCH=ON flag during compilation (see above) must not be used if this software is used for other purposes than research. This flag automatically downloads, compiles and links against multi-label graph cut optimization which

can be used only for research purposes. For commercial purposes, be aware that there is a US patent on the main algorithm itself.

Cited from the multi-label graph cut optimization README file. See that file for further information.

If you use our texturing code for research purposes, please cite our paper:

@inproceedings{Waechter2014Texturing,
  title    = {Let There Be Color! --- {L}arge-Scale Texturing of {3D} Reconstructions},
  author   = {Waechter, Michael and Moehrle, Nils and Goesele, Michael},
  booktitle= {Proceedings of the European Conference on Computer Vision},
  year     = {2014},
  publisher= {Springer},
}

Contact

If you have trouble compiling or using this software, if you found a bug or if you have an important feature request, please use the issue tracker of github: https://github.com/nmoehrle/mvs-texturing

For further questions you may contact us at mvs-texturing(at)gris.informatik.tu-darmstadt.de