Project for Shapenet Segmentation Challenge, 2017
This is an implementation of the KD-Tree network method used for the challenge in Keras.
Uses inspiration from Escape from Cells: Deep Kd-Networks for the Recognition of 3D Point Cloud Models for pre-processing the data. The architecture is based on the famous U-Net
Official report available here
The data provided by the organizers must be extracted into a folder 'data'.
- prepare_data.py - processes data and packages them into numpy arrays
- model.py - defines and trains the model
- generate_segs.py - to generate labels for test/validation models post training