3D object classification using the ModelNet40 dataset
This project is a Jupyter Notebook where I worked on building and training a point cloud classification model using deep learning techniques. The notebook walks through the entire pipeline — from data loading and preprocessing to training and evaluating the model.
- Dataset Handling: Code to load, normalize, and prepare 3D point cloud datasets for training and testing.
- Data Augmentation: Includes jittering, random rotations, and scaling to make the model robust.
- Model Architecture: A neural network model inspired by PointNet, which includes input and feature transformation networks, followed by global feature extraction and fully connected layers.
- Training Process: The model is compiled and trained using Keras, with accuracy and loss tracked across epochs.
- Saving & Visualization: The trained model is saved, and training history is visualized with accuracy/loss plots. There’s also a section for visualizing predictions on test data.
This project helped me understand the flow of working with 3D data, applying augmentation techniques, and building a deep learning pipeline specifically for point cloud classification tasks.
Just a personal learning project — not meant for production use.