smaranjitghose / movenet

Fastest Pose Estimation on any device

This repo comprises of a script that helps us to detect 17 keypoints from the human body and draw the corresponding edges for Pose Detection.

MODEL USED : MoveNet Lightning SinglePose

Usage ⭐ :

• Clone this repository

git clone www.github.com/smaranjitghose/movenet

• Move inside the cloned directory of the project

cd movenet

• Make sure Python is installed and set to PATH

• Install the dependencies [Creating a virtual Environment may be useful beforehand]

pip install - requirements.txt

• Run the script

python main.py

Note:

• For Linux/Mac based systems, replace the python with python3 in the above command(s)
• In order to get a live demonstration, open the main.py script in a text editor or IDE like VS Code and change the parameter passed to the cv2.VideoCapture() function to 0 or 1 to obtain the feed in real time from your camera
• In order to try this on other videos, repeat the above step and change the update the variable input_video_path.
• For ensuring efficacy in particular scenarios (which might be dependent on factors like lightning, skin color, background light and age), try tweaking the threashold value
• Remember, this is solely for prototyping purposes. For better results, use the Thunder Version.

References/Credits:

TFHUB MoveNet

movenet's Issues

Nose, eyes and ears are mispositioned

Great job!
Nose, eyes and ears are mispositioned
How to correct it please?