Bayesian Optimization of Pointnet

Hyperparameter optimization using bayesian optimization on the popular Pointnet framework based on the works of Revisiting Small Batch Training for Deep Neural Networks by Dominic Masters, Carlo Luschi and Algorithms for hyper-parameter optimization by James Bergstra, R´emi Bardenet, Yoshua Bengio, Bal´azs K´egl

Prerequisites

• Git
• Anaconda 1.9.2
• Jupyter Notebook 5.5.0
• Spyder 4.0
• Python 3.6

Libraries

• hyperopt
• tensorflow
• tensorboard

Setup

All scripts can be run directly on google colab.

Running the solution

Please run the following commands on a terminal.

• git clone https://github.com/steve7an/PointnetEnhanced.git
• jupyter notebook
• Open Pointnet_Training_and_Evaluate.ipynb on notebook
• run each section independently to test Pointnet, Pointnet++ or 3DmFV

Hyperopt is currently only enabled for Pointnet.