Home

Awesome

totrilib - Torch Transient Imaging Library

Official implementation of "Fast Differentiable Transient Rendering for Non-Line-of-Sight Reconstruction" [paper]

Install

Totrilib in only tested on Linux, but should also work on Windows.

Docker

Make sure you have Docker and NVIDIA Container Toolkit installed. Download NVIDIA Optix (7.2.0) from https://developer.nvidia.com/designworks/optix/downloads/legacy and extract to extern/NVIDIA-OptiX-SDK-7.2.0, and make sure you have pybind11 in extern/pybind11. The easiest way to run totrilib is to use Docker Compose and start a detached container with docker compose up -d. Otherwise see docker-compose.yml for settings to start a container. Mount additional volumes as needed.

Standard

Make sure you have the following dependencies installed. The suggested versions should work fine, but others might also do.

Install with pip (pip install /path/to/totrilib).

Samples

Code will follow soon.

To run the sample code, various data files are required. For the Statue, Diffuse S, Mannequin and the Zaragoza Bunny we refer to the respective publications listed below. Adjust the paths in the sample files as needed.

Citation

If you find this repository helpful please cite the following paper.

  @inproceedings{plack2023fast,
    title={Fast Differentiable Transient Rendering for Non-Line-of-Sight Reconstruction},
    author={Plack, Markus and Callenberg, Clara and Schneider, Monika and Hullin, Matthias B},
    booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
    pages={3067--3076},
    year={2023}
  }