“RightAnkle”, | # 0 |
“RightKnee”, | # 1 |
“RightHip”, | # 2 |
“LeftHip”, | # 3 |
“LeftKnee”, | # 4 |
“LeftAnkle”, | # 5 |
“RightWrist”, | # 6 |
“RightElbow”, | # 7 |
“RightShoulder”, | # 8 |
“LeftShoulder”, | # 9 |
“LeftElbow”, | # 10 |
“LeftWrist”, | # 11 |
“Nose”, | # 12 |
“RightEye”, | # 13 |
“LeftEye”, | # 14 |
“RightEar”, | # 15 |
“LeftEar”, | # 16 |
{
'poses': [frame1_poses, frame2_poses, frame3_poses, ...],
'confidence': [frame1_confidences, frame2_confidences, frame3_confidences, ...],
'header': {
'task': 'pose_estimation',
'joints_index':{
'RightAnkle':0,
'RightKnee':1,
'RightHip':2,
'LeftHip':3,
'LeftKnee':4,
'LeftAnkle':5,
'RightWrist':6,
'RightElbow':7,
'RightShoulder':8,
'LeftShoulder':9,
'LeftElbow':10,
'LeftWrist':11,
'Nose':12,
'RightEye':13,
'LeftEye':14,
'RightEar':15,
'LeftEar':16
}
}
}
}
Where ‘poses’ and ‘confidence’ are of the following structure:
Empty frameN_confidences and frameN_poses means there are no persons detected on frame.
Order of joints is reflected in ‘header.joints_index’ mapping.
Full structure looks like:
{
'poses': [ [ [ [x1,y1], [x2,y2], ...], [ [x1,y1], [x2,y2], ...], ...], [ [ [x1,y1], [x2,y2], ...], ...], ...],
'confidence': [ [ [c1, c2, ...], [c1, c2, ...], ...], [ [c1, c2, ...], ...], ...],
'header': {
'task': 'pose_estimation',
'joints_index':{
'RightAnkle':0,
'RightKnee':1,
'RightHip':2,
'LeftHip':3,
'LeftKnee':4,
'LeftAnkle':5,
'RightWrist':6,
'RightElbow':7,
'RightShoulder':8,
'LeftShoulder':9,
'LeftElbow':10,
'LeftWrist':11,
'Nose':12,
'RightEye':13,
'LeftEye':14,
'RightEar':15,
'LeftEar':16
}
}
}
}