Classifying point clouds of different objects using PointNet
We will be focussing on classifying point clouds as belonging to one of several different object classes.
Shown below is the architecture proposed by the authors of the PointNet paper. This implementation will take up the classification network alone.
Download the dataset from the website given here:- http://3dvision.princeton.edu/projects/2014/3DShapeNets/
You can use either the Model10 or Model40 dataset, make sure you change the folder path to the dataset in the config.py file accordingly.
To run the training process on the downloaded dataset:- python pointnet.py
All hyperparameters such as learning rate, number of epochs, batch size,
number of classes in the dataset, number of points in the pointcloud can be set in the config.py file
PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation