Pointnet/pointnet++ training and testing

1、 install Anaconda
source ~/.bashrc Update environment variables
Then you can type conda list Test success

2、 establish PyTorch Environmental Science
conda create -n myPytorch python=3.7
conda activate mypytorch

conda install pytorch1.1.0 torchvision0.3.0 -c pytorch

3、 install cloudcompare Software
snap install cloudcompare

install meshlab Software
meshlab2020.07-linux.AppImage

4、 Copy pointnet project
git clone github website

5、 Complete the shape classification task
Download datasets ：modelnet40_normal_resampled
Decompress it and put it in data Under the folder

python train_cls.py --model pointnet2_cls_msg --normal --log_dir pointnet2_cls_msg

Use normal vector information

If the video memory overflows , Set up batch_size

python train_cls.py --model pointnet2_cls_msg --normal --log_dir pointnet2_cls_msg batch_size 8

Test the trained network

python test_cls.py --normal --log_dir pointnet2_cls_msg

6、 Object component segmentation
Using data sets ：shapenetcore_partanno_segmentation_benchmark_v0_normal
Unzip the dataset to data Folder
Training orders ：

python train_partseg.py --model pointnet2_part_seg_msg --normal --log_dir pointnet2_part_seg_msg

The test command ：

python test_partseg.py --normal --log_dir pointnet2_part_seg_msg

7、 Scene semantic segmentation

Data sets ：Stanford3dDataset_v1.2_Aligned_Version
Unzip to ：
data/s3dis/Stanford3dDataset_v1.2_Aligned_Version/

Training ：

cd data_utils
python collect_indoor3d_data.py

The processed data is saved to

data/stanford_indoor3d/

Execute training orders ：

python train_semseg.py --model pointnet2_sem_seg --test_area 5 --log_dir
pointnet2_sem_seg

The visualization results are saved in

log/sem_seg/pointnet2_sem_seg/visual/

Execute test command ：

python test_semseg.py --log_dir pointnet2_sem_seg --test_area 5 --visual