FMCW_Doppler_PointTransformerNet

This repository reproduces DopplerPTNet.
The codebase is an unofficial implementation of DopplerPTNet for semantic segmentation.
This code is created based on the descroption of DopplerPTNet. If you find the implementation is not accurate, please raise an issue.

---

Dependencies

• Ubuntu: 20.04 or higher

• PyTorch: 1.9.0

• CUDA: 11.1

• Hardware: Batch size 16 requires 48GB memory.

• To create conda environment, command as follows:

  bash env_setup.sh dopplerpt
  

Dataset preparation

• No public dataset for doppler lidar point cloud yet.

• Download S3DIS dataset and symlink the paths to them as follows:

  mkdir -p dataset
  ln -s /path_to_s3dis_dataset dataset/s3dis
  

Usage

• Shape classification on ModelNet40
  • For now, please use paconv-codebase branch.
• Part segmentation on ShapeNetPart
  • For now, please use paconv-codebase branch.
• Semantic segmantation on S3DIS Area 5

  • Train

    • Specify the gpu used in config and then do training:

      python train.py --config config/s3dis/s3dis_pointtransformer_repro.yaml
      

  • Test

    • Afer training, you can test the checkpoint as follows:

      CUDA_VISIBLE_DEVICES=0 python test.py config/s3dis/s3dis_pointtransformer_repro.yaml
      

  ---

Experimental Results

---

TODO

• Semantic segmentation training and testing code
• Object detection training code
• Experiment result on non-FMCW LIDAR
• Experiment result on FMCW LIDAR
• CUDA Deployment code

---

References

If you use this code, please cite:

@software{DopplerPointTransformerNet-pytorch,
  author = {Dingkang Wang},
  month = {12},
  title = {{DopplerPointTransformerNet-pytorch}},
  url = {https://github.com/dingkwang/DopplerPTNet-pytorch},
  version = {0.1},
  year = {2024}
}

Acknowledgement

The description is from DopplerPTNet: Object Detection Network with Doppler Velocity Information for FMCW LiDAR Point Cloud. I also refer point-transformer.