/PAPNet

Code for "Classification of Single-View Object Point Clouds"

Primary LanguagePythonMIT LicenseMIT

PAPNet

Code for "Classification of Single-View Object Point Clouds"

Requirements

This code has been tested with

  • python 3.7.12
  • pytorch 1.6.0
  • CUDA 10.1
  • se3cnn [link]

Downloads

  • PartialModelNet40(PM40) dataset [link]
  • PartialScanNet15(PS15) dataset [link]
  • Pretrained models of PAPNet [link]
  • PartialModelNet40(PM40) under different levels of partiality [link]

Training

Command for training PAPNet:

python train.py --dataset pm40 --data_path path_to_pm40_dataset
python train.py --dataset ps15 --data_path path_to_ps15_dataset

Evaluation

Evaluate the results of PAPNet:

python test.py --dataset [pm40|ps15] --data_path path_to_pm40_or_ps15 --model path_to_pretrained_model
python transfer_test_pm2ps.py --pm40_model path_to_pm40_model --ps15_path path_to_ps15_dataset
python transfer_test_ps2pm.py --ps15_model path_to_ps15_model --pm40_path path_to_pm40_dataset

License

Our code is released under MIT License (see LICENSE file for details).