PyTorch code for FSDCNet

A Pytorch implementation of "Fusion of a Static and Dynamic Convolutional Neural Network for Multiview 3D Point Cloud Classification"

More information about the paper is in here

Dependencies

• Blender 2.92.0
• Python 3.6
• Cuda 10.1
• PyTorch 1.7.1
• torchvision 0.8.2
• Open3d
• Pandas
• numpy
• matplotlib

First, download ModelNet40 and Sydney Urban Object dataset and use preprocessing.py to convert all samples to pcd format files.

Second, use blender to render it to 2D Images. You can refer to the tutorial and our codes render.py.

Third, when you get png files of these multi-view pointcloud, put it in specified directory and execute the following command:

python train_FSDCNet.py -name FSDCNet -num_models 1000 -lr 0.0001 -weight_decay 0.0001 -num_views 6 -bs 16 -cnn_name dy_resnet50

If you find FSDCNet useful in your research, please consider citing. Many thanks!

Wang, W.; Zhou, H.; Chen, G.; Wang, X. Fusion of a Static and Dynamic Convolutional Neural Network for Multiview 3D Point Cloud Classification. Remote Sens. 2022, 14, 1996. https://doi.org/10.3390/rs14091996

@article{wang2022fusion,
  title={Fusion of a Static and Dynamic Convolutional Neural Network for Multiview 3D Point Cloud Classification},
  author={Wang, Wenju and Zhou, Haoran and Chen, Gang and Wang, Xiaolin},
  journal={Remote Sensing},
  volume={14},
  number={9},
  pages={1996},
  year={2022},
  publisher={MDPI}
}