What is in it
Dataset · datatorch/coco-person-keypoints-val2017 · 607 files
COCO Person Keypoints (val2017, open licenses)
Person keypoint subset of COCO 2017 val with openly licensed images only: 605 images whose Flickr license is CC BY 2.0 (413), CC BY-SA 2.0 (189) or No known copyright restrictions (3), with 1,609 people annotated with the 17 COCO keypoints (nose, eyes, ears, shoulders, elbows, wrists, hips, knees, ankles), COCO visibility flags and bounding boxes. Every image is credited with its Flickr URL and license in its file metadata and in annotations/credits.csv. License: annotations CC BY 4.0 (COCO Consortium); each image keeps its Flickr owner's license. Source: https://cocodataset.org Please cite: Lin, T.-Y. et al., Microsoft COCO: Common Objects in Context. ECCV 2014.
Open →
| Fact | Value |
|---|---|
| Source | COCO 2017 val, person keypoints |
| Images | 605 |
| People annotated | 1,609 |
| Per person | 17 keypoints with visibility flags, and a bounding box |
| Image licenses | CC BY 2.0 (413), CC BY-SA 2.0 (189), No known copyright restrictions (3) |
| Annotation license | CC BY 4.0, COCO Consortium |
| Credits | every image's Flickr URL and license, in annotations/credits.csv and in each file's metadata |
The 17 keypoints are the standard COCO set: nose, eyes, ears, shoulders, elbows, wrists, hips, knees and ankles. Their order, the skeleton pairs and the visibility rules are in the COCO keypoint format reference.
Why only openly licensed images
COCO's annotations are CC BY 4.0, but the images are Flickr photos that keep their owners' licenses, and many of those don't allow commercial use or adaptations. If you want to show training examples in a paper, a demo or a product page, that matters.
This subset keeps only the images whose license allows reuse with credit, and credits every one. It's small enough to page through in an afternoon and large enough to fine-tune or sanity-check a pose model.
How to use it
- 1
Browse it
Open the dataset above and open any image. Each person's skeleton is drawn on top, with hidden points marked as not visible. No download needed.
- 2
Download it
Use Download dataset on the project page. A free account is needed. You get the images and the COCO keypoint JSON as a ZIP, or you can pull it with the DataTorch Python client (see the project page).
- 3
Label your own images the same way
Use the same 17-point skeleton on your own photos so your data and this set line up. See how to annotate keypoints in DataTorch.
Good uses
- Fine-tune or check a pose model on images you're allowed to show in results.
- Test a training pipeline end to end on real COCO data without downloading all of COCO.
- Learn the format. Compare the JSON with the drawn skeletons to see exactly what each visibility flag means.
- Convert to another layout. The keypoint formats comparison has code to go from COCO 17 to MPII, MediaPipe or COCO-WholeBody.
Source and license
COCO 2017 val, person keypoints. Annotations: CC BY 4.0, COCO Consortium. Images: each keeps its Flickr owner's license (CC BY 2.0, CC BY-SA 2.0 or No known copyright restrictions), listed per image in annotations/credits.csv. Source: https://cocodataset.org. Please cite: Lin, T.-Y. et al., Microsoft COCO: Common Objects in Context, ECCV 2014.
Label keypoints on your own images
Use the same 17-point skeleton in the browser, with your team.
Frequently asked questions
How many images are in this COCO keypoints subset?
605 images with 1,609 annotated people, all from COCO 2017 val.
Why not all of COCO val2017?
COCO's images keep their Flickr owners' licenses, and many of those are non-commercial or no-derivatives. This subset keeps only images licensed CC BY 2.0 (413), CC BY-SA 2.0 (189) or No known copyright restrictions (3), so it can be shown and shared with credit.
Can I use it commercially?
The licenses kept here allow reuse with attribution, including commercial reuse, and CC BY-SA asks that adaptations be shared under the same license. Each image's license and Flickr URL are in annotations/credits.csv and in the file metadata. Check them for your use.
What format are the annotations in?
COCO keypoints. Each person has 17 keypoints as x, y and a visibility flag (0 not labeled, 1 labeled but not visible, 2 visible), plus a bounding box.
Open the hosted copy on DataTorch.
Browse the files in your browser, then download the dataset with a free account.
