Copyright disclaimer

The copyright in this software is being made available under the BSD LICENSE Copyright (c) 2017-present Leibniz University Hannover (LUH) Institut fuer Informationsverarbeitung (TNT)

The code provided here was adapted from:

@inproceedings{kluger2017deep,
  title={Deep learning for vanishing point detection using an inverse gnomonic projection},
  author={Kluger, Florian and Ackermann, Hanno and Yang, Michael Ying and Rosenhahn, Bodo},
  booktitle={German Conference on Pattern Recognition (GCPR)},
  year={2017}
}

The paper can be found on arXiv. Refer to fkluger/vanishing_points_2017 for additional information and resources.

• Caffe 1.0-RC5 on Python 2.7
• ImageMagick
• Python requirements.txt packages

• Launch the docker container from https://hub.docker.com/repository/docker/davidelanz/vanishing_points

  docker pull davidelanz/vanishing_points
  

• Test the software:

  cd /home/vanishing_points
  pytest test.py
  

• Launch a docker container from https://hub.docker.com/r/bvlc/caffe (Python 2.7 + Caffe)

  docker pull bvlc/caffe:cpu
  

  or any other suitable tag, e.g.:

  docker pull bvlc/caffe:gpu
  

• Initialize the repository

  git clone https://github.com/Davidelanz/vanishing_points.git
  cd vanishing_points
  pip install -r requirements.txt
  

• Download the CNN weights and image mean files from the releases and put them into the cnn folder.
• Adjust config.py so that it contains the path to your Caffe installation and the paths where you store the benchmark datasets.

You can run the vanishing point detector on the example images and visualise the results. Computation may take a few moments. Adjust the gpu_id in the main.py file if necessary. Then run:

python main.py

A single-image test is available:

pytest test.py