Skip to content

vinits5/PointNetLK

Repository files navigation

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

  1. Train PointNetLK:

python main.py

  1. Test PointNetLK:

python test.py

  1. 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

About

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

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages