Forum

OpenCV/Open Pose Human Pose Recognition

#1

Hello Everybody!

I am relatively new to VVVV and gamma and I have a question about the
optimization regarding OpenCV/Open Pose Human Pose Recognition.

Is there a way to make it faster and more real-time? Because currently
it is very laggy and loses frames.

I know that this should work because I have had experience with PoseNet in P5.js
and there the performance was much faster so I was really surprised why
VVVV was so slow with Human Pose recognition.

I am working on a university project, which should put static, stylized human clothes
on humans in real-time.

So basically I was wondering how to make the performance better. and then later
to put low poly meshes onto the recognized human.

Could someone please help? :) Is there any path I can follow, documentation, meet up, etc? :)

#2

Hi @mishamatrix, I am afraid the library we are using for VL.OpenCV works on the CPU only which for tasks like openPose is not fast enough for realtime.

What you have previously tested is likely GPU powered.

I am currently working on bringing GPU support to VL.OpenCV but this will take some time.

For now I would suggest to look for other alternatives that make use of the GPU such as https://github.com/takuya-takeuchi/OpenPoseDotNet.

Cheers.

#3

@ravazquez Thank you very much!
I am very excited about the VL.OpenCV GPU support!

Thank you for the inforimation, I will check it out!

All the best to you!

Cheers

© vvvv group