3D classification primitive with PyTorch Geometric

Description

This project demonstrates application of PyTorch Geometric and other technologies for creating a simple demo app.

Repository with model here

Used technologies:

• Voila for creating web-application
• Jupyter Widgets for GUI
• BinderHub for running

Voila theme used from other repository

Requirements

1. Python 3.7 or higher.
2. Anaconda or Miniconda.
3. Git LFS (weights for model is storing in LFS).

How to tun

Create anaconda environment:

conda env create -n 3d-cls-demo --file ./environment.yml
conda activate 3d-cls-demo

For running in Voila:

python ./setup.py develop

Files from directory share must be in right place, otherwise will be error when Voila starting. For Anaconda:

jupyter --paths

Possible output for Windows:

data:
    D:\...\3d-cls-demo\share\jupyter

Directory structure in D:\...\3d-cls-demo\share\jupyter must be:

├───nbconvert
│   └───templates
│       ├───asciidoc
│       ├───base
│       ├───basic
│       ├───classic
│       │   └───static
│       ├───compatibility
│       ├───html
│       ├───lab
│       │   └───static
│       ├───latex
│       ├───markdown
│       ├───material
│       │   └───static
│       ├───python
│       ├───reveal
│       │   └───static
│       ├───rst
│       └───script
└───voila
    └───templates
        ├───base
        │   └───static
        ├───classic
        │   └───static
        ├───lab
        └───material
            └───templates

Note material name. This is directory with settings of theme for Voila application. Corresponding directories must be copy manually. Some problems were with setup.py under Windows.

After you need to invoke command:

voila ./demo.ipynb --template material