Created by Pedro Hermosilla, Sebastian Maisch, Tobias Ritschel, Timo Ropinski.

This repository contains the code of our EGSR paper, Deep-learning the Latent Space of Light Transport. A video of our method can be found in the following link.

If you find this code useful please consider citing us:

@article{hermosilla2018ginn,
    title={Deep-learning the Latent Space of Light Transport},
    author={Pedro Hermosilla and Sebastian Maisch and Tobias Ritschel and Timo Ropinski },
    journal={Computer Graphics Forum (Proc. EGSR 2019)}
}

numpy
pygame
CUDA 9.0
tensorflow 1.12

The first step is downloading the code for Monte Carlo Convolutions (MCCNN) in the following link. The software expects the code to be in a folder named MCCNN. The second step is following the instruction on the README from MCCNN to compile the library.

Modify the compiling script in folder rt_viewer/cuda_ops with your cuda and python3 paths. Then, execute the compiling script to generate the CUDA/OpenGL operations. Lastly execute the scripts rt_viewer/sss.sh or rt_viewer/gi.sh to use the trained networks to visualize two 3D models.

In order to train a network on our datasets first download the data from the following link (COMMING SOON). Then, execute the script processData.py to generate the numpy files. Lastly, execute the command:

python GITrainRT.py --useDropOut --useDropOutConv --augment --dataset 2

The parameter dataset determines which effect the network should be trained on.