Lightweight pose estimation with Mediapipe

Ever wondered how do those fancy whatsapp filters work? Here I will use holistic mediapipe for creating full-body meshing.

Written by: @tornikeo and @GPT3

The embeddable iframe allows any casual user to use pose estimation within their web-app, for free!

Overview by @tornikeo

Pose estimation is the process of determining the position and orientation of an object in an image. It can be used for a variety of purposes, such as 3D reconstruction, object tracking, and human-computer interaction. The MediaPipe technology can be used to estimate pose from the camera input in real-time. This web-app is completely free, is self-hosted and is very fast.

Additional benefits include - the ability to simply copy-paste the iframe tag to include the pose estimation on any website. The open source nature of this project (here’s the full repo).

Some limitations would be as follows: Since this is self-hosted, its performance heavily depends on the client-side compute capability. Clients without GPU access, for example, could experience drop in FPS.

Future work would be to create android or iOS app for pose estimation - potential uses could be estimating athletic performance, calculating calories spent on exercise or, use AR (augmented reality) approaches to modify the predicted pose into interesting shapes for fun!

Overview by @GPT3

Potential use-cases for pose estimation web-app:

Pose estimation can be used for a variety of applications. Some potential use-cases for a pose estimation web-app include:

  • Tracking the movement of objects in a 3D space
  • Generating 3D models of objects
  • Creating virtual reality experiences
  • Analyzing human movement for health or fitness purposes

Usefulness of the web-app:

The web-app is completely free, is self-hosted and is very fast (real-time). This makes it an ideal tool for developers or businesses who want to create applications or services that make use of pose estimation.

Summary and future goals:

The pose estimation web-app is a simple and fast way to estimate pose from camera input. The app is free and self-hosted, making it easy to use for developers or businesses. The goal is to eventually port the app to Android or iOS, making it even more accessible.