This is re-implementation of PC-NBV in pytorch. Official TensorFlow implementation: pcnbv_tf
- Python 3.6.10
- PyTorch 1.2.0
- CUDA 10.1
- Follow this guide to install Open3D for point cloud I/O.
- Clone the repository:
git clone xxxx
- Data Preparation Please follow the official repo pcnbv_tf to generate NBV data.
- Train
python train.py
- Evaluate the model:
If you want to show AUC results for each class, please run:
python Manager.py python cal_cov_shapenet.py python drawAUC_compare_shapenet.py
python cal_AUC.py
Class | pytorch | tensorflow |
---|---|---|
Airplane | 0.798 | 0.799 |
Cabinet | 0.609 | 0.612 |
Car | 0.614 | 0.612 |
Chair | 0.784 | 0.782 |
Lamp | 0.802 | 0.800 |
Sofa | 0.642 | 0.640 |
Table | 0.763 | 0.760 |
Vessel | 0.721 | 0.719 |
Bus | 0.672 | 0.677 |
Bed | 0.662 | 0.662 |
BookShelf | 0.738 | 0.740 |
Bench | 0.847 | 0.845 |
Guitar | 0.844 | 0.849 |
Motorbike | 0.730 | 0.728 |
Skateboard | 0.844 | 0.840 |
Pistol | 0.670 | 0.672 |
Average | 0.734 | 0.733 |