Using PointNet on Tunnel Dataset

This repo is implementation for PointNet

Environment

The latest codes are tested on Ubuntu 16.04, CUDA10.1, PyTorch 1.6 and Python 3.7
conda install pytorch==1.6.0 cudatoolkit=10.1 -c pytorch

Classification(Tunnel Dataset)

Use your own tunnel dataset 4-channel
Like this picture (x , y, z, intensity)

Data save in data/tunnel_data/normal data/tunnel_data/defect

Performance

Model	Input Point	Batch Size	Defect Accuracy	Normal Accuracy	Instance Accuracy
PointNet-3ch	1024	12	68.0	56.0	62.0
PointNet-3ch	1024	24	64.0	80.0	72.0
PointNet-3ch	2048	12	60.0	88.0	74.0
PointNet-3ch	2048	24	64.0	88.0	76.0
PointNet-4ch	1024	12	84.62	87.5	86.0
PointNet-4ch	1024	24	88.46	91.67	90.0
PointNet-4ch	2048	12	92.31	79.17	86.0
PointNet-4ch	2048	24	88.46	75.0	82.0

Visualization Result

Point cloud ground truth ( Red:defect, White:normal ) Prediction result

Reference

Pointnet_Pointnet2_pytorch