keras-io/PointNet

Model Description

Keras Implementation of Point cloud classification with PointNet

This repo contains the trained model of Point cloud classification with PointNet.

The full credit goes to: David Griffiths

Intended uses & limitations

• As stated in the paper, PointNet is 3D perception model, applying deep learning to point clouds for object classification and scene semantic segmentation.
• PointNet takes raw point cloud data as input, which is typically collected from either a lidar or radar sensor.

Training and evaluation data

• The dataset used for training is ModelNet10, the smaller 10 class version of the ModelNet40 dataset.

Training procedure

Training hyperparameter

The following hyperparameters were used during training:

• optimizer: 'adam'
• loss: 'sparse_categorical_crossentropy'
• epochs: 20
• batch_size: 32
• learning_rate: 0.001

Model Plot

View Model Plot