the real coordinates of the images

[blahole]

I've inferred several images and found that the gestures are quite similar, but the rotation angle of the wrist joint is not consistent with the images. I would like to ask whether the predicted 21 keypoints can be mapped to the real coordinates of the images in the model design.