An unofficial PoinetNet

This is an unofficial PointNet with Tensorflow2.

Performance on ModelNet40 classification dataset

set	Accuracy
test	88+

Performance on ShapeNet part segmentation dataset

mIOU	Airplane	Bag	Cap	Car	Chair	Earphone	Guitar	Knife	Lamp	Laptop	Motorbike	Mug	Pistol	Rocket	Skateboard	Table
83.5	82.0	82.7	82.3	75.5	89.7	69.3	91.3	86.0	80.6	94.7	66.0	91.7	82.2	54.2	72.3	80.6

Performance on Saturn multi-label classification dataset

set	Accuracy
test	100

Performance on Saturn part segmentation dataset

set	Accuracy
test	100

My Environment

• Operating System:
  • Archlinux
• CUDA:
  • CUDA V11.5.50
• Nvidia driver:
  • 495.44
• Python:
  • python 3.6.5
• Python package:
  • h5py...
• Tensorflow:
  • tensorflow-2.6.0

Downloading the ModelNet40 Classification Dataset

ModelNet40

Downloading the Pretrain Model

Model

For training

cd ./Classification

python main.py

For testing

cd ./Classification

python main.py --phase='test'

For visualization

cd ./Classification

python main.py --phase='test' --vis=1

Downloading the ShapeNet Part Segmentation Dataset

ShapeNet-Org ShapeNet-h5

Downloading the Pretrain Model

Model

For training

cd ./PartSeg

python main.py

For testing (sample)

cd ./PartSeg

python main.py --phase='test'

For testing (original)

cd ./PartSeg

python test_all.py

For visualization

cd ./PartSeg

python main.py --phase='test' --vis=1

python test_all.py --vis=1

Downloading the Saturn Multi-label Classification Dataset

Saturn

Downloading the Pretrain Model

Model

For training

cd ./Toy/Classification

python main.py

For testing

cd ./Toy/Classification

python main.py --phase='test'

For visualization

cd ./Toy/Classification

python main.py --phase='test' --vis=1

Downloading the Saturn Part Segmentation Dataset

Saturn

Downloading the Pretrain Model

Model

For training

cd ./Toy/PartSeg

python main.py

For testing

cd ./Toy/PartSeg

python main.py --phase='test'

For visualization

cd ./Toy/PartSeg

python main.py --phase='test' --vis=1

Part Segmentation Sample Results from Saturn and ShapeNet

References

1. charlesq34/pointnet

2. Charles R. Qi, Hao Su, Kaichun Mo, and Leonidas J. Guibas. 2017b. PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation. In Proc. CVPR.