README.md

Graph Representations to Model Physical Systems
Explore the docs »

View Demo · Project Report

---

Bachelor project 2020 - Machine Learning - USI

Advisor: Prof. Cesare Alippi

Co-advisors: Andrea Cini & Daniele Zambon

Table of Contents

• About the Project
  • Built With
• Getting Started
  • Prerequisites
  • Installation
• Usage
• License
• Contact

About The Project

The following project consists in a python package that applies 2D multi-person pose estimation to images and videos, and parses the results into attributed graphs. The goal of this project is to make available an all-in-one tool to build datasets based on real time 2D multi-person detection, and is part of my bachelor project at USI.

The software works with images, videos and real time input streams. It uses OpenPose to parse the given frames into keypoints and generates JSON files, containing a clear and rich representation of attributed graphs for each detected person.

It also provides some utilities for converting the data from a JSON structured graph, to a more standard representation such as the NetworkX format. For more info 👉🏼click here.

Built With

• Python
• OpenPose
• Anaconda
• Networkx
• Spektral
• MPII Human Pose Dataset
• Matlab

Getting Started

Prerequisites

a) Python should be installed on your system 👉🏼 Python download link.

b) Install OpenPose and its dependencies 👉🏼 prerequisites and installation guide

c) Make sure to export pyOpenPose 👉🏼API doc

Installation

1. Clone GraphPipe from github:

git clone https://github.com/gabecarra/GraphPipe.git
cd GraphPipe

2. Install package dependencies

python setup.py install

3. Make sure that pyOpenPose works:

python
import openpose

4. Done! 🎉🥳

Usage

GraphPipe can be used both via command line interface or as a python package:

• CLI

python graph_pipe [ARGS]

• python package

import graph_pipe
graph_list = graph_pipe.run([ARGS])

Examples:

• Body pose estimation from a video

python graph_pipe  --video examples/media/video.avi --write_json output/ --display 0

• Body + hands + face from images

python graph_pipe --image_dir examples/media/ --face --hand

For more examples, please refer to the OpenPose doc

For the complete set of FLAGS see flags.hpp

License

Distributed under the MIT License. See LICENSE for more information.

Logo icons made by Becris and itim2101 from www.flaticon.com

Contact

Gabriel Henrique Carraretto - [email protected]

Project Link: https://github.com/gabecarra/GraphPipe