Keypoint and pose estimation datasets.
Where to get keypoint-annotated data: datasets hosted on DataTorch you can download directly, and the public benchmarks every pose paper reports on: what they cover, how big they are, and how many points they annotate.
Download directly
Community keypoint datasets on the DataTorch hub: browse the annotations in place, download as a ZIP with a free account.
Animated Drawings
1,300+ files
Character drawings annotated with pose keypoints in COCO format: the humanoid skeleton applied to hand-drawn figures.
View datasetASI Sun Location
1,800+ files
All-sky imager frames with the sun position annotated as a point: single-landmark localization at scale (NREL SRRL, Golden, Colorado).
View datasetMore on the hub: all public keypoint datasets →
Public pose and landmark datasets
Sizes are the widely cited approximate figures. Licenses differ (several are research-only), so check the dataset’s own terms before commercial use.
| Dataset | Subject | Scale | Keypoints | Notes |
|---|---|---|---|---|
| COCO Keypoints | People | ~200k images, ~250k person instances | 17 body keypoints | The de-facto standard; defines the format most tools exchange. |
| MPII Human Pose | People | ~25k images, ~40k people | 16 body joints | Everyday activity coverage; the classic single-person benchmark. |
| CrowdPose | People (crowded) | ~20k images, ~80k persons | 14 keypoints | Built for occlusion and overlapping people. |
| AP-10K | Animals | ~10k images, 50+ species | 17 keypoints | The broadest animal-pose benchmark. |
| FreiHAND | Hands | ~130k hand images | 21 hand keypoints | 3D-annotated hands; the standard hand-pose starting point. |
| WFLW | Faces | ~10k faces | 98 facial landmarks | Dense facial landmarks with occlusion/expression attributes. |
Exchanging between these? Most tooling speaks the COCO keypoint format: the full format reference covers the order, skeleton, and visibility flags. New to the annotation type itself? Start with the keypoint annotation guide.
Need keypoints on your own imagery?
Define the skeleton once, annotate point-by-point in the browser with guided ordering and visibility flags, and export COCO keypoints, or host the finished dataset publicly on the hub, like the ones above.
The annotation platform for specialized imagery: review, score, and share datasets your team works from.
© 2026 DataTorch. All rights reserved.