This repository is for "Focal Loss in 3D Object Detection". If you use this code for academic purpose, we would appreciate if you cite our paper:

@article{yun2018fl3d,
  title={Focal Loss in 3D Object Detection},
  author={Peng Yun, Lei Tai, Yuan Wang, and Ming Liu},
  journal={arXiv preprint arXiv:1809.06065},
  year={2018}
}

It contains two parts: 3D-FCN and VoxelNet as claimed in our paper. For detailed information of these two parts, please refer the sub-README in each dir.

The splited data & weights are available in Data&Weights

The docker images for 3D-FCN and VoxelNet are also available.

docker pull pyun/voxelnet:baseline
docker pull pyun/3d-fcn:tf1.7
# Then launch a container with nvidia-docker 1.0

• Kitti evaluation borrows from prclibo/kitti_eval.
• Network skeletons are adapted from
  • qianguih/voxelnet
  • jeasinema/VoxelNet-tensorflow
  • yukitsuji/3D_CNN_tensorflow.