About plot_3axis_Zaxis

[Mt-Taka0129]

Hello, I have questions about demos/image.py.

def plot_3axis_Zaxis(img, yaw, pitch, roll, tdx=None, tdy=None, size=50., limited=True, thickness=2):
    # Input is a cv2 image
    # pose_params: (pitch, yaw, roll, tdx, tdy)
    # Where (tdx, tdy) is the translation of the face.
    # For pose we have [pitch yaw roll tdx tdy tdz scale_factor]

    p = pitch * np.pi / 180
    y = -(yaw * np.pi / 180)
    r = roll * np.pi / 180

When drawing the three axes, why is only the Yaw angle multiplied by a minus sign?

Also, is the head posture, the output of the neural network, an Eulerian angular representation (z-x-y system) with the camera coordinate system (image below) rotated in the order of Z-axis, Y-axis and X-axis?

[hnuzhy]

This is adopted from the original implementation of HopeNet and FSA-Net.

Please refer links below:
https://github.com/natanielruiz/deep-head-pose/blob/master/code/utils.py#L51
https://github.com/shamangary/FSA-Net/blob/master/demo/demo_FSANET.py#L18

For the order of three Euler angles, we used the Z-Y-X mode as in 300W-LP and AFLW2000. Actually, I recently found that the inherent drawbacks of Euler angles (e.g., the gimbal lock, and sensitive to rotation order around three axes) can be addressed by the rotation matrix representation. I will turn to it and also suggest you to use it.