Shih-Hung Liu, Shang Yi Yu, Shao-Chi Wu, Hwann-Tzong Chen, Tyng-Luh Liu
ubuntu 16.04 + cuda 10.1
python 3.6
pytorch 1.5.1
scipy 1.3
h5py 2.9
open3d-python 0.3.0
S3DIS: we use the same data released by JSIS3D. You can download the data into the ./data_s3dis
ScnaNet: you can download the ScanNet data in ScanNet.
python train.py
python main_eval.py
- Compiling the pointnet++ module
cd Pointnet2.PyTorch/pointnet2
python setup.py install
- You also need to compiling SCN for semantic prediction
The environment is based on facebookresearch/SparseConvNet
The pretrained GICN on S3dis dataset is in ./experiment
Evaluation on Area5:
-precision : 0.6348
-recall : 0.4669
Pointnet++ is based on sshaoshuai/Pointnet2.PyTorch