/PC-NBV_pytorch

Implementation of PC-NBV in Pytorch.

Primary LanguagePython

PC-NBV: A Point Cloud Based Deep Network for Efficient Next Best View Planning

Introduction

This is re-implementation of PC-NBV in pytorch. Official TensorFlow implementation: pcnbv_tf

Environment

  • Python 3.6.10
  • PyTorch 1.2.0
  • CUDA 10.1
  • Follow this guide to install Open3D for point cloud I/O.

Installation

  1. Clone the repository: 
    git clone xxxx
  2. Data Preparation Please follow the official repo pcnbv_tf to generate NBV data.
  3. Train 
    python train.py
  4. Evaluate the model: 
    python Manager.py
python cal_cov_shapenet.py
python drawAUC_compare_shapenet.py
    If you want to show AUC results for each class, please run: 
    python cal_AUC.py

Result

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