README.md

Master Project of 3D-CNN

This is the master project built for recognizing objects within point clouds.

Point cloud data classification built with Tensorflow.

This project contains two parts which is concept and application respectively. While the concept contains all the trail python scripts along with many Jupyter Notebooks, the application is built with Angular and Python Flask.

Thesis is under /doc folder

Web app for CNN model

Requirements

• node
• python 3
• Training data. Sample training data can be found here.

Dependencies

Running with Jupyter Notebook only.

• Anaconda Python 3
• PyntCloud (Unlicensed)
• tensorflow

Additional packages for starting the app.

• npm
• @angular/cli
• flask

Dependencies For Running on GPU

Tensorflow-GPU version requires Nvidia GPU which supports CUDA. You can find the list of valid GPUs here.

• tensorflow-gpu
• CUDA

After Installation, you can try running this script to verify you got all dependencies on your Windows machine. You may find it under '/tools/tensorflow-self-check.py'.

Usage

• Home Page

• Training

• Segmentation

• Prediction

• Layer Extraction

How to run

For running the app:

Before running:

$ cd ./app/angular
$ npm install

For running:

$ bash run.bash

For cleaning up:

$ bash clean.bash

For running the concept:

$ cd ./concept
$ jupyter notebook