Shape from Caustics

Official PyTorch implementation of the main ideas described in our paper Shape from Caustics: Reconstruction of 3D-Printed Glass from Simulated Caustic Images.

Prerequisites

NVidia OptiX 6.5 Make sure the shared libraries are in PATH or LD_LIBRARY_PATH respectively.
PyTorch >= 1.7
PyWavefront
PyTorch Wavelets (for using wavelet sparsity)
PyWavelets
Matplotlib
tqdm
Tensorboard
PyMongeAmpere (for running the reimplementation of High-contrast computational caustic design)

Setup

First build the necessary OptiX .ptx files; we have provided a CMakeLists.txt file for this task (which should also be automatically invoked, when executing setup.py). To get the files into the correct location, the install Target has to be called:

mkdir build && cd build
cmake ..
cmake --build . --target install

Second build the PyTorch extensions by invoking python setup.py install. Be sure to change the paths in setup.py to the correct OptiX directory.

Executing the code

If everything is set up correctly, you can call python shape_from_caustics.py --help to get an overview of the parameters for simulation and reconstruction. Alternatively you can also look at hyperparameter_helper.py to see, which parameters are available and which might take a list of arguments. A call of python shape_from_caustics.py will start the simulation and reconstruction of a synthetic 3D printed glass sample with sensible initial parameters (for a GPU with 24GB VRAM). If you have problems with Out of memory errors, try decreasing the num_inner_simulations parameter. For a complete overview of parameters for the result in the paper, have a look at the supplementary material.

If you want to execute the reimplementation of High-contrast computational caustic design, look at schwartzburg_2014/ma.py and change the paths therein to the respective paths of PyMongeAmpere and cgal-python as well as to the input and output images in __main__.

Citation

If you use this code for your publications, please cite our paper using the following BibTeX.

@inproceedings{kassubeck2020shape,
  title = {Shape from Caustics: Reconstruction of 3D-Printed Glass from Simulated Caustic Images},
  author = {Kassubeck, Marc and B{\"u}rgel, Florian and Castillo, Susana  and Stiller, Sebastian and Magnor, Marcus},
  booktitle = {{IEEE}/{CVF} Winter Conference on Applications of Computer Vision ({WACV})},
  pages = {2877--2886},
  month = {Jan},
  year = {2021}
}

Acknowledgements

The authors would like to gratefully acknowledge funding from the German Science Foundation (DFG) under Germany’s Excellence Strategy within the Cluster of Excellence PhoenixD (EXC 2122, Project ID 390833453), and from the German Federal Ministry of Education and Research (grant No. 05M18MBA-MOReNet).