yellowshuohahaha / pointnetlk

This is the clean implementation of PointNetLK with pretrained models. (A 3D point cloud registration network)

PointNetLK: Point Cloud Registration using PointNet

Source Code Author: Vinit Sarode

This code is tested using torch 1.3.0+cu92 and torchvision 0.4.1+cu92.

Code use:

1. Train PointNet:

python train_classifier.py

2. Train PointNetLK:

python main.py

3. Test PointNetLK:

python test.py

4. Test PointNetLK (with sample data):

python test_sample.py

Pretrained Models:

1. best_model: PointNetLK pretrained model.
2. best_ptnet_model: PointNet_features model.
3. classifier: PointNet classification model.

Requirements:

1. PyTorch
2. ModelNet40 [Link]
3. SciPy
4. GPU Requirements: Approximately 6 GB of GPU memory for batch size = 16.

References:

1. PointNetLK [paper]
2. PointNetLK [github]
3. PCRNet [github]
4. DCP [github]

Sample Results:

Template: Red Point Cloud, Source: Green Point Cloud & Registration Result: Blue Point Cloud