Point Cloud Classification Model based on Dual-Input Deep Network Framework

Data

You need to download the data to the “data“ folder and unzip it before training the model.
RELATED LINKS：

ModelNet40:"https://shapenet.cs.stanford.edu/media/modelnet40_ply_hdf5_2048.zip"
ScanObjectNN:"https://github.com/hkust-vgd/scanobjectnn"

Usage

The code has been tested with Python 3.6.9, TensorFlow-gpu 1.12.0, on windows10.

• To train DI-PointNet on ModelNet40 run
  python train.py --data_set ModelNet40 --weight 10 --err_data 5000
  in "DI-PointNet".
• To train DI-PointNet on ScanObjectNN run
  python train.py --data_set ScanObjectNN --weight 100 --err_data 3000
  in "DI-PointNet".
• To train DI-PointCNN on ModelNet40 run
  python train_val_cls.py --setting ModelNet_x3_l4 --weight 0.1 --err_data 4921
  in "DI-PointCNN".
• To train DI-PointCNN on ScanObjectNN run
  python train_val_cls.py --setting ScanObjectNN_x3_l4 --weight 10 --err_data 5708
  in "DI-PointCNN".

References

• PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation by Qi et al. (CVPR 2017).
• PointCNN: Convolution On X-Transformed Points by Li et al. (NIPS 2018).