Regarding yaw, pitch and roll in test_own_images.ipynb about img2pose HOT 4 CLOSED

Thanks @vujadeyoon

Yes, the pose_pred includes rotation vectors (left axis rotation, top axis rotation, front axis rotation) and translation coordinates.
If you want pitch, yaw and roll (in this order) on the same format as AFLW2000 dataset, you can use the conversion below:

def convert_to_aflw(rotvec, is_rotvec=True):
    if is_rotvec:
        rotvec = Rotation.from_rotvec(rotvec).as_matrix()
    rot_mat_2 = np.transpose(rotvec)
    angle = Rotation.from_matrix(rot_mat_2).as_euler('xyz', degrees=True)
    
    return np.array([angle[0], -angle[1], -angle[2]])

However, this format suffers from a problem where yaw is limited to (-90, 90) degrees, and if go beyond this point, pitch and roll are affected to compensate.

Otherwise, just used the first three values from the pose_pred which corresponds to the rotation vector.

Hope this helps.

from img2pose.

Ok I got your point. Thank you for letting me know.

from img2pose.

Dear 976jay,

You can find code [1] that you want in the jupyter cell, Run on a folder or an image list.
I think the variable, pose_pred, includes information related to the yaw, pitch, and roll angles as well as 3D translation coordinate values (i.e. x, y, z)

Best,
vujadeyoon

[1] pose_pred = res["dofs"].cpu().numpy()[i].astype('float')

from img2pose.

I will look into this. Thank you so much

from img2pose.