SLAM based on deep learning odometry for camera position estimation.

1. Share results from paper "Training Deep SLAM on Single Frames" [arxiv].
2. Share framework for training, evaluating and storing results of various odometry models.

• libqglviewer-dev-qt4
• libeigen3-dev
• libsuitesparse-dev
• conda
• rest of requirements listed in conda.yml

git clone --recursive https://github.sec.samsung.net/AIMC-TSU/odometry.git

1. conda env update --file conda.yml
   
2. conda activate odometry
3. bash build.sh
   Or on cluster, you must
   a) ssh to any gpu node
   b) scl enable devtoolset-7 'bash build.sh'
4. Update PYTHONPATH: export PYTHONPATH=path_to_this_repo:$PYTHONPATH
5. (optional) For depth estimation download pretrained struct2depth weights from official site https://sites.google.com/view/struct2depth into weights directory
6. (optional) For OF estimation download pretrained tfoptflow weights from repo https://github.com/philferriere/tfoptflow

python -m unittest discover -s tests

The code is released under the MPL 2.0 License. MPL is a copyleft license that is easy to comply with. You must make the source code for any of your changes available under MPL, but you can combine the MPL software with proprietary code, as long as you keep the MPL code in separate files.

If you use this code for your research, please site our paper:

@misc{slinko2019training,
    title={Training Deep SLAM on Single Frames},
    author={Igor Slinko and Anna Vorontsova and Dmitry Zhukov and Olga Barinova and Anton Konushin},
    year={2019},
    eprint={1912.05405},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}