3DCFS: Fast and Robust Joint 3D Semantic-Instance Segmentation via Coupled Feature Selection
The illustration of CFSM seen as follow
The details of our proposed 3DCFS architecture seen as follow
The results of our method on S3DIS Area5 and 6-Fold CV respectively.
1.clone the 3DCFS repository:
cd ~
git clone https://github.com/Biotan/3DCFS
2.download the dataset such as S3DIS and modify the path in train and test file.
3.prepare the environment including tensorflow1.2.0 and python2.7.
-
make the code of pointnet according to [Pointnet].
-
trainning
cd 3DCFS/models
python train.py
- testing and evaluation
python test.py && python eval_iou_accuracy.py