COCO keypoint visualizer: pose, check, export.
The 17 COCO person keypoints as a live skeleton. Drag points, flip visibility flags, paste an annotation to see it rendered with every format mistake named, and copy a valid annotation back out. Runs in your browser; nothing is uploaded.
Need the reference instead? The COCO keypoint format lists the order, skeleton and flags.
Four things you can do here
Drag a point
Every circle is one of the 17 COCO person keypoints, numbered with its 1-based skeleton id. Drag it and the skeleton edges follow; the export on the right updates as you move.
Click to set visibility
Select a point and choose v = 2 (visible), 1 (labeled but occluded) or 0 (not labeled). Occluded points draw in blue with dashed edges; unlabeled points are hollow and export as 0, 0, 0.
Paste an annotation
Drop in a single annotation, a whole COCO JSON file, or a bare array of 51 numbers. The figure re-poses to match and every format problem is named by index: wrong length, a bad visibility flag, a mismatched num_keypoints, a category whose keypoint names are out of order, a 0-based skeleton.
Copy the export
The right-hand panel is a valid COCO annotation for the pose on screen, with num_keypoints and bbox computed. Useful for unit-test fixtures, seeding a dataset, or checking what a correct file looks like.
Array index is 0-based (keypoints[0..16]); skeleton ids are 1-based (1..17). The figure shows ids, the inspector shows both.
Common questions#
Does this upload my data anywhere?
No. Parsing, rendering and export all run in your browser. Nothing is sent to a server.
Why does my left shoulder show on the right?
COCO sides are the subject’s, not the viewer’s. In a front-facing image the subject’s left shoulder is on the viewer’s right. The checker warns when the shoulders look swapped, since mirrored exports are a common bug.
Which errors does the checker catch?
Wrong keypoints length (not 51), non-numeric values, visibility flags other than 0, 1 or 2, unlabeled points with non-zero coordinates, num_keypoints that does not match the count of v > 0, category keypoint names out of the standard order, and a skeleton that uses 0-based ids.
Can it show more than one person, or another category?
Not yet. It renders the first annotation in a file that has keypoints, and it knows the 17-point person schema only. For real multi-person, multi-category work use the DataTorch annotator, which reads and writes the same format.
Want to annotate real images this way?
The DataTorch annotator does this on your photos: define the keypoints and skeleton on a label, place points in order with a live overlay, mark occlusion, and import or export COCO keypoints unchanged.
Tools and community for image annotation: label, review, and share datasets, from everyday photos to the imagery only specialists can read.
© 2026 DataTorch. All rights reserved.