Garrick Brazil, Xiaoming Liu

Monocular 3D region proposal network source code as detailed in arXiv report, accepted to ICCV 2019 (Oral). Please also visit our project page.

Our framework is implemented and tested with Ubuntu 16.04, CUDA 8.0, Python 3, NVIDIA 1080 Ti GPU. Unless otherwise stated the below scripts and instructions assume working directory is the project root.

If you utilize this framework, please cite our ICCV 2019 paper.

@inproceedings{brazil2019m3drpn,
    title={M3D-RPN: Monocular 3D Region Proposal Network for Object Detection},
    author={Brazil, Garrick and Liu, Xiaoming},
    booktitle={Proceedings of the IEEE International Conference on Computer Vision},
    address={Seoul, South Korea},
    year={2019}
}

• Cuda & Python

  In this project we utilize Pytorch with Python 3, Cuda 8, and a few Anaconda packages. Please review and follow this installation guide. However, feel free to try alternative versions or modes of installation.

• Data

  Download the full KITTI detection dataset. Then place a softlink (or the actual data) in M3D-RPN/data/kitti.

   cd M3D-RPN
   ln -s /path/to/kitti data/kitti
  

  Then use the following scripts to extract the data splits, which use softlinks to the above directory for efficient storage.

  python data/kitti_split1/setup_split.py
  python data/kitti_split2/setup_split.py
  

  Next, build the KITTI devkit eval for each split.

   sh data/kitti_split1/devkit/cpp/build.sh
   sh data/kitti_split2/devkit/cpp/build.sh
  

  Lastly, build the nms modules

  cd lib/nms
  make
  

We use visdom for visualization and graphs. Optionally, start the server by command line

python -m visdom.server -port 8100 -readonly

The port can be customized in scripts/config files. The training monitor can be viewed at http://localhost:8100

Training is split into a warmup and main configurations. Review the configurations in scripts/config for details.

python scripts/test.py --config=kitti_3d_multi_front_alpha

If your training is accidentally stopped, you can resume at a checkpoint based on the snapshot with the restore flag. For example to resume training starting at iteration 10k, use the following command.

python scripts/test.py --config=kitti_3d_multi_front_alpha --restore=10000

For questions regarding M3D-RPN, feel free to post here or directly contact the authors {brazilga, liuxm}@msu.edu.