PPFNet-tensorflow

Unoficial implementation of PFFNet for pointcloud class.

Project architecture:

• src/
  • dataset.py : handles the data preprocessing, constructing the patches.
  • utils.py : Save point cloud to ply files, to open with cloud compare.
  • threeDMatch.py : reads the SUN3D dataset as used in 3DMatch, compute normals and position of the points.
  • shapenet.py : Reads the shapenet dataset, sample points on the 3D models.
  • loss.py : implementation of the N-tuple loss.
  • pointnet.py : pointnet of the PPFNet model.

Notbooks:

• dataset-visualization.ipynb : Visualize the dataset, testing that it is well constructed
• dataset.ipynb : preprocess the pairs of the SUN3D dataset and test the speed of the shapenet dataset
• main-shapenet.ipynb : train PPFNet on shapenet
• main.ipynb : train PPFNet on SUN3D
• model-visualization.ipynb : Visualize the result of the trained models

Dataset used:

• shapenet
• SUN3D from 3DMatch

Frameworks:

• tensorflow 2
• jupyter
• trimesh (to load shapenet)
• plyfile (to export ply files)
• pandas (Reading shapenet split)