convolutional-pose-machines-tensorflow icon indicating copy to clipboard operation
convolutional-pose-machines-tensorflow copied to clipboard

How can two hands be recognized and posed

Open momo1986 opened this issue 6 years ago • 3 comments

For example, I have one previous picture.

image It is recognized as: image When it cut with two separated hands, OK: image

image It looks like that the numbers of the joints is set as 21, one hand will have almost 20 points. If multiple hands, e.g., 2 hands have, how can I recognize it successfully? Will the setting of joint-number work?

momo1986 avatar Jan 30 '19 07:01 momo1986

Can any update be given?

momo1986 avatar Feb 12 '19 02:02 momo1986

Hi, The model should take single hand image as input. You can use a hand detector (ex. YOLO) to detect each hand, crop hands and feed into model one by one.

timctho avatar Feb 12 '19 10:02 timctho

This model can only predict right hand, however, I have tried this model with horizontally flipped image. The result was a surprise, it actually can predict left hand too, but you need to flip the predicted result.

What you need to do are:

  1. feed original image
  2. feed flipped image
  3. flip it back as left hand prediction

Zen3515 avatar Feb 14 '19 07:02 Zen3515