Lite-FPN for Keypoint-based Monocular 3D Object Detection

This is the official implementation of our manuscript Lite-FPN for Keypoint-based Monocular 3D Object Detection.

Requirements

All codes are tested under the following environment:

• Ubuntu 18.04
• Python 3.7
• Pytorch 1.3.1
• CUDA 10.1

Dataset

We train and test our model on official KITTI 3D Object Dataset. Please first download the dataset and organize it as following structure:

kitti
│──training
│    ├──calib 
│    ├──label_2 
│    ├──image_2
│    └──ImageSets
└──testing
     ├──calib 
     ├──image_2
     └──ImageSets

Install

1. We use conda to manage the environment:

conda create -n Lite-FPN python=3.7

conda install pytorch=1.3 torchvision -c pytorch
conda install yacs scikit-image tqdm numba fire pybind11

pip install mmcv-full==1.2.5
pip install mmdet==2.11.0

git clone https://github.com/open-mmlab/mmdetection3d.git
cd mmdetection3d
git checkout v0.9.0
pip install -v -e .  # or "python setup.py develop"

2. Build codes:

cd Lite-FPN
python setup.py build develop

3. Link to dataset directory:

mkdir datasets
ln -s /path_to_kitti_dataset datasets/kitti

Getting started

First check the config file under configs/.

Training :

python tools/plain_train_net.py --config-file "configs/smoke_gn_vector.yaml"

Evaluation :

python tools/evaluate_script.py --config-file "configs/smoke_gn_vector.yaml"

Citation

If you find our work useful in your research, please consider citing:

@article{Yang2021LiteFPNFK,
  title={Lite-FPN for Keypoint-based Monocular 3D Object Detection},
  author={Lei Yang and Xinyu Zhang and Li Wang and Minghan Zhu and Jun Li},
  journal={ArXiv},
  year={2021},
  volume={abs/2105.00268}
}

Acknowledgements

Many thanks to these excellent open source projects: SMOKE

Contact

If you have any problem with this code, please feel free to contact yanglei20@mails.tsinghua.edu.cn.