Skip to content

prashanthnetizen/posenet_nodejs_setup

Repository files navigation

Pose Estimation using PoseNet

Pre-Requisites:

  • Python3
  • OpenCV Python version
  • Node js
  • Tensorflow for js
  • tensorflow-models

Installation (Windows 10):

  • Download the Node JS installer for windows from here
  • After installing Node js, add it to your Environment PATH variable.
  • Please make sure whether you have installed Microsoft .NET Framework 4.5.1+ in your windows machine.
  • Install Windows Build Tools globally to support running node js by using the following command:
npm install --global --production windows-build-tools
  • Unzip the code in your machine locally and do the following inside the directory to install necessary packages locally.
npm install
  • Use the same command to install any missing packages mentioning its name.

Executing:

  • Get your recorded videos and place them all inside a directory. Note down the path to the directory as master path.
  • You can use the python script -> Frames_Extractor.py and specify the master path to the variable "path_to_video_files" and run it to obtain some further directories inside your path containing video frames of respective videos.
  • After installing node js server and required libraries, use the js file "scale_to_videos.js" and specify the master path to the variable "photo_path_to_frames" and execute this script.
  • The console asks for your choice of Architecture for PoseNet -> Residual Net gives better accuracy compared to Mobile Net but the latter is fast.
  • After providing your choice of architecture, wait for key points to get generated for individual videos inside their corresponding directories.
  • Optional -> You can use the convert_to_csv.py script to convert the key_points.json files to CSV files.

"Intelligence is the ability to avoid doing work, yet getting the work done." - Linus Torvalds


Cheers ! Get Inspired :)

About

For Generating Key Points using Tensorflow's PoseNet

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published