0

I am using openpose Tensorflow for multi personal pose estimation. But as it follows COCO 18 keypoints detection, it is taking lots of time to detect. How can i reduce this detection of keypoints to detect only one part(eg: leg). means I don't want to detect full body. instead, I want to detect only legs with reduced keypoint numbers to increase the fps

I followed this tutorial:https://towardsdatascience.com/realtime-multiple-person-2d-pose-estimation-using-tensorflow2-x-93e4c156d45f

But I am getting an fps around 3. since it is detecting full-body, in my case it is a waste process, becoz 1 only want to detect one part(eg: leg) to increase the fps

Scott
  • 4,974
  • 6
  • 35
  • 62
Firos Vp
  • 1
  • 1

1 Answers1

0

For a beginner of deep learning I will say there is not a easy way to revise it to detect only a single part.

If you are familiar with this it you can change the pafprocess remove unused parts and recompile it, but actually I'm not sure about if you should retrain the model or not.

But I thought to use a lighter model is easier than change the code.

SheiUn
  • 9
  • 1
  • 4