A CNN model trained on the ModelNet 40 dataset.
About the dataset:
The Princeton ModelNet40 contains 3D models in the form of .off files.This implementation uses data in the form of .h5py files, which can be found here.
How it works:
- Point cloud data is read from the .h5py files.
- It is then voxelised to reduce dimensionality.
- The voxelised array is fed as input to the model defined in train.py.
- Set the path to save the model in train.py.
- Use test.py to visualise misclassified examples.
Requirements
- Tensorflow 1.x
- Numpy
- Open3D
- Matplotlib