ankitwahane497 / fl3d

The code is implemented to show focal loss improvement based on 3D-FCN and VoxelNet for “Focal Loss in 3D Object Detection”.

README

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}
}

Organization

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.

Data & Weights

The splited data & weights are available in Data&Weights

Docker

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

Acknowledgments

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