This update to Ultralytics' JSON2YOLO converter, which initially supported only segmentation tasks, builds upon Ryouchinsa's converter and contributions from other developers.

While Ryouchinsa’s version enabled pose detection, it was limited to 17 keypoints. This update now allows for the conversion of COCO WholeBody .json files, which contain 133 keypoints, into YOLO format, for more comprehensive pose detection and further processing. It may be adjusted for variable keypoints as well.

def set_coco_keypoints(use_keypoints, w, h, ann, box, keypoints, type="default"):
    if not use_keypoints:
        return
    if 'keypoints' not in ann:
        keypoints.append([])
        return
    if len(ann['keypoints']) == 0:
        keypoints.append([])
        return
    else:
        if type == "default":
            k = (np.array(ann['keypoints']).reshape(-1, 3) /
                 np.array([w, h, 1])).reshape(-1).tolist()
            k = box + k
            keypoints.append(k)
        elif type == "foot":
            k = (np.array(ann['foot_kpts']).reshape(-1, 3) /
                 np.array([w, h, 1])).reshape(-1).tolist()
            keypoints.append(k)
        elif type == "face":
            k = (np.array(ann['face_kpts']).reshape(-1, 3) /
                 np.array([w, h, 1])).reshape(-1).tolist()
            keypoints.append(k)
        elif type == "lefthand":
            k = (np.array(ann['lefthand_kpts']).reshape(-1, 3) /
                 np.array([w, h, 1])).reshape(-1).tolist()
            keypoints.append(k)
        elif type == "righthand":
            k = (np.array(ann['righthand_kpts']).reshape(-1, 3) /
                 np.array([w, h, 1])).reshape(-1).tolist()
            keypoints.append(k)

Update by Selim Gilon on Nov 13

Simply added the option to export 17, 23, or 133 keypoints based on the user's needs. 17 can be too limiting while 133 might be too complex and unnecessary.

• 17 = body
• 23 = body + feet
• 133 = body + feet + hands + face

def convert_coco_json():
# rest of the code
                if num_keypoints >= 17:
                    set_coco_keypoints(use_keypoints, w, h,
                                       ann, box, body_keypoints)
                if num_keypoints >= 23:
                    set_coco_keypoints(use_keypoints, w, h,
                                       ann, box, foot_keypoints, type="foot")
                if num_keypoints == 133:
                    set_coco_keypoints(use_keypoints, w, h,
                                       ann, box, face_keypoints, type="face")
                    set_coco_keypoints(use_keypoints, w, h,
                                       ann, box, lefthand_keypoints, type="lefthand")
                    set_coco_keypoints(use_keypoints, w, h,
                                       ann, box, righthand_keypoints, type="righthand")
# rest of the code

Example usage:

• COCO .json file

• 

• Converted YOLO .txt file

Each .txt annotation file in YOLO format contains either 56, 74, or 404 values, structured as follows:

1	2	3	4	5	Keypoints ([17, 23, 133] x 3)
class-id	x_center	y_center	width_bb	height_bb	<x_p1> <y_p1> <v_p1> <x_p2> <y_p2> <v_p2> ... <x_pn> <y_pn> <v_pn>

where n = [17, 23, 133]

Example when n=133 :

0 0.609594 0.513106 0.341719 0.815718 0.573438 0.190588 2 0.584375 0.171765 2 0.5625 0.176471 2 0.603125 0.183529 2 0.55625 0.190588 2 0.623437 0.254118 2 0.559375 0.303529 2 0.676562 0.334118 2 0.532813 0.374118 2 0.701562 0.388235 2 0.482812 0.418824 2 0.6625 0.477647 2 0.614062 0.503529 2 0.670312 0.691765 2 0.573438 0.642353 2 0.728125 0.851765 2 0.61875 0.802353 2 0.685937 0.889412 2 0.696875 0.894118 2 0.748437 0.870588 2 0.589063 0.844706 2 0.5875 0.842353 2 0.645312 0.830588 2 0.555973 0.177318 1 0.556803 0.186079 1 0.558194 0.194676 1 0.560184 0.202993 1 0.562728 0.21096 1 0.566041 0.218244 1 0.570319 0.224288 1 0.575761 0.227401 1 0.581548 0.226216 1 0.586944 0.222784 1 0.591564 0.217333 1 0.59547 0.210744 1 0.598286 0.202998 1 0.599838 0.194473 1 0.600117 0.18565 1 0.600156 0.176802 1 0.599848 0.167967 1 0.557944 0.171704 1 0.560316 0.168905 1 0.563156 0.167869 1 0.566214 0.167427 1 0.569312 0.167503 1 0.576517 0.16588 1 0.579602 0.164277 1 0.582814 0.163187 1 0.586156 0.16298 1 0.589622 0.164439 1 0.572766 0.174021 1 0.573022 0.179176 1 0.57325 0.18424 1 0.573422 0.189402 1 0.570475 0.194256 1 0.572356 0.194901 1 0.574445 0.194832 1 0.576831 0.193426 1 0.579208 0.192123 1 0.560477 0.177539 1 0.562959 0.17433 1 0.566213 0.173747 1 0.569222 0.175726 1 0.566473 0.177807 1 0.563456 0.178561 1 0.577767 0.173584 1 0.580327 0.170032 1 0.583773 0.169295 1 0.587066 0.171093 1 0.5842 0.173409 1 0.580992 0.174174 1 0.567475 0.204545 1 0.569981 0.201055 1 0.57347 0.199406 1 0.575075 0.198909 1 0.576692 0.19852 1 0.581536 0.197772 1 0.586067 0.199741 1 0.583886 0.204879 1 0.580694 0.20897 1 0.5766 0.211055 1 0.572922 0.210833 1 0.569877 0.208485 1 0.568106 0.204478 1 0.571442 0.201999 1 0.575288 0.200674 1 0.580381 0.199599 1 0.585386 0.200107 1 0.581509 0.205392 1 0.576266 0.207951 1 0.57182 0.207821 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.475162 0.42765 1 0.469847 0.43006 1 0.464531 0.432471 1 0.460469 0.438824 1 0.453281 0.441882 1 0.454531 0.415529 1 0.449219 0.414118 1 0.446094 0.419765 1 0.4475 0.429176 1 0.45125 0.422118 1 0.445312 0.425882 1 0.448906 0.437882 1 0.455937 0.445882 1 0.449531 0.429882 1 0.443438 0.433176 1 0.447656 0.444941 1 0.453125 0.451765 1 0.447969 0.436 1 0.441875 0.440941 1 0.445 0.450824 1 0.450625 0.457647 1