FastPCI: Motion-Structure Guided Fast Point Cloud Frame Interpolation (ECCV 2024)

Tianyu Zhang, Guocheng Qian, Jin Xie and Jian Yang

Requirements

Create a conda environment:

conda env create -f environment.yaml
conda activate fastpci

cd models/EMD/
python setup.py install
cp build/lib.linux-x86_64-cpython-39/emd_cuda.cpython-39-x86_64-linux-gnu.so .

cd ../pointnet2/
python setup.py install

cd ../../

Dataset preparation

We utilize the NL-Drive dataset, which processed and integrated KITTI Odometry, Argoverse2sensor, and Nuscenes. Please download the NL-Drive dataset here . And put the NL-Drive dataset into data/NL-Drive . We provide the split list of three datasets in ./data/NL-Drive/.

Instructions to training and testing

Training

Training on KITTI Odometry dataset, Argoverse 2 sensor dataset, Nuscenes dataset, run separately:

bash train_fastpci_kitti.sh
bash train_fastpci_argoverse2.sh
bash train_fastpci_nuscenes.sh

Testing

Testing on KITTI Odometry dataset, Argoverse 2 sensor dataset, Nuscenes dataset, run separately:

bash test_fastpci_kitti.sh
bash test_fastpci_argoverse2.sh
bash test_fastpci_nuscenes.sh

Citation

If you find our code or paper useful, please cite:

@inproceedings{zhang2024fastpci,
  title     = {FastPCI: Motion-Structure Guided Fast Point Cloud Frame Interpolation},
  author    = {Zhang, Tianyu and Qian, Guocheng and Xie, Jin and Yang, Jian},
  booktitle = {Proceedings of the European Conference on Computer Vision (ECCV)},
  year      = {2024}
  }

Acknowledgments

We thank the authors of

• PointPWC-Net
• PointINet
• NeuralPCI
• EMA-VFI

for open sourcing their methods.