This is a tensorflow implementation of pointnet based on the official implementation. Main difference is the thorough replacement of conv2d operations by tf.layers.dense.
To train the classification model:
python train.py
ModelNet40 models in HDF5 files will be downloaded to ./data. Training results, including model parameter, hyperparameter setting and log files, will be saved into ./training_results/<training time>.
You can tune hyperparameters in the base_config.ini file. Each time you train the model, this file will be read and copied into ./training_results/<training time>.