Keypoint & pose annotation services
Keypoint annotation marks a fixed, ordered set of landmarks on each object — such as the 17 COCO body joints from nose to ankles — with a visibility flag for every point. Connected into a skeleton, the keypoints train pose estimation models that track posture, movement and orientation.
- Trains
- Pose estimation
- From
- $0.091 / label
- Review
- Every job
- Formats
- COCO keypoints JSON
Is it the right label?
When to use keypoints & pose
Pose estimation — the model predicts where each joint or landmark is on every person or object in a new image.
You care about posture or movement, not just presence: sports technique, ergonomics, fall detection, fitness.
You need precise landmark positions: facial landmarks, hand joints, animal pose, object corners.
Downstream logic depends on geometry between points — angles, distances, symmetry.
When something else fits better
If you only need to know a person is there, keypoints are overkill — boxes are faster. If the subject is mostly hidden in most frames, agree how invisible joints are handled first or the labels become guesses.
Bounding boxesHow we label it
Our process
- 01
Fix the skeleton template
Point names, order and connections are defined once — COCO-17 or your own — and locked so every annotation uses the same order.
- 02
Per-point visibility
Every joint is marked visible, occluded or not in frame. Hidden joints are flagged, not guessed into position.
- 03
Left/right discipline
Left and right are the subject's, not the viewer's. The flip mapping is recorded so mirror augmentation does not poison training.
- 04
Skeleton review
Reviewers check point placement and left/right swaps, the most common pose error, before approval.
Quality
What our QA checks
Every job is reviewed by a person other than the annotator, on top of automated checks. Work below the agreed threshold goes back for rework.
Per-joint placement error against reviewer corrections
Left/right swap detection
Visibility-flag consistency on occluded joints
Point-order and template validation on every instance
Pricing & delivery
Rates and formats
| Annotation | Rate |
|---|---|
| Keypoints & pose | $0.091 / label |
QA review is included. Bulk discounts apply to large projects; the final quote is confirmed after we review a sample of your data. Full pricing
Delivered as
Used for
Sports analytics · Fitness and physiotherapy apps · Workplace ergonomics · Fall detection · Animal behaviour · Facial landmarks
Industries
Industries that use keypoints & pose
Keypoints & pose FAQ
Keypoints & pose, answered
What is keypoint annotation?
Keypoint annotation places a fixed, ordered set of points — body joints, facial landmarks, object corners — on each instance in an image, with a visibility flag for each point. It is the training data for pose estimation.
What are the 17 COCO keypoints?
Nose, left and right eye, left and right ear, left and right shoulder, elbow, wrist, hip, knee and ankle. It is the most widely used human-pose template; we also label custom skeletons.
How are hidden joints labeled?
Each keypoint carries a visibility flag: visible, occluded but inferable, or not present. Occluded joints are placed only when the guideline allows inference; otherwise they are marked absent rather than guessed.
What is flip_idx and why does it matter?
flip_idx tells a training pipeline which point becomes which when an image is mirrored — left wrist becomes right wrist. If it is wrong, horizontal-flip augmentation silently teaches the model swapped limbs. We deliver the mapping with every pose dataset.
Keep reading
Related services and guides
Service
Bounding boxes
Tight 2D boxes around every object, class by class.
Service
Video & object tracking
Frame-by-frame tracks with persistent IDs.
Service
Polygons & instance segmentation
Pixel-accurate outlines and masks per object.
Guide
COCO vs YOLO vs Pascal VOC: annotation formats explained
Guide
How to write data annotation guidelines that annotators follow
Guide
How to measure annotation quality: IoU, agreement and gold sets
Need keypoints & pose on your data?
Send us a few sample images or clips. We'll agree the guideline with you and send a plan and a per-label quote.