pictograph/coco-2017-pose

View on Pictograph · Pictograph Research · Custom License

About

COCO 2017 Pose is a computer-vision dataset from COCO 2017, hosted on Pictograph. The most common detected objects are person, child, ski, crowd, skateboard, racket. On Pictograph you can browse every annotated image, import it into your own workspace in one click, export it in a dozen formats, or train a model on it directly.

At a glance

Metric	Value
Images	58,931
Keypoints	1,709,789
Skeleton	person, 17 keypoints
Annotation types	keypoint
Splits	train, validation

Quick start

Load it in one line with the datasets library, then read each record's COCO-pose keypoint instances:

from datasets import load_dataset

ds = load_dataset("pictograph/coco-2017-pose", split="train")
example = ds[0]
image = example["image"]                  # a PIL image

# one entry per object instance (COCO-pose)
for inst in example["keypoints"]:
    names = inst["node_names"]            # e.g. ["nose", "left_eye", "right_eye", ...]
    flat = inst["keypoints"]              # [x, y, v, x, y, v, ...] aligned to names
    joints = {
        names[i]: (flat[3 * i], flat[3 * i + 1], flat[3 * i + 2])
        for i in range(len(names))
    }
    print(inst["category"], joints)       # v: 0 unlabelled, 1 occluded, 2 visible

Prefer a full annotation editor, one-click import, multi-format export, and one-click training? Open this dataset on Pictograph.

Dataset structure

This dataset uses the Hugging Face imagefolder layout: each split directory holds the images plus a metadata.jsonl that links every image to its annotations by file_name.

Field	Description
file_name	Path to the image within the split directory.
objects.bbox	Bounding boxes as [x, y, width, height] (pixels).
objects.categories	Integer class index per box (matches the class list below).
objects.category_names	Human class name per box.
keypoints	Keypoint instances (COCO-pose): object category plus a flat [x, y, v, ...] array aligned to node_names (the skeleton below); v is 0 unlabelled, 1 occluded, 2 visible. Each instance's box in objects encloses its keypoints.

Data instance

One record (each keypoints entry is one object instance; keypoints is a flat [x, y, v, ...] array of 17 keypoints aligned to node_names, v: 0 unlabelled, 1 occluded, 2 visible):

{
  "image": <PIL.Image (RGB)>,
  "objects": {"bbox": [[172.0, 96.0, 140.0, 310.0]], "categories": [0], "category_names": ["person"]},
  "keypoints": [
    {
      "category": "person", "category_id": 0, "num_keypoints": 17,
      "node_names": ["nose", "left_eye", "right_eye", ...],
      "keypoints": [201.0, 148.0, 2.0, 188.0, 146.0, 1.0, 0.0, 0.0, 0.0, ...]
    }
  ]
}

Classes

Class index matches objects.categories and each keypoint instance's category_id.

#	Class	Keypoints
0	person	17

Skeleton

Each object is one instance: its keypoints share an instance_id and are named for their joint. A keypoint that is labelled but hidden carries occluded: true. The joints are in this order in every export (a COCO keypoint array follows it).

#	Keypoint	Connects to
0	nose	left_eye, right_eye
1	left_eye	right_eye, nose, left_ear
2	right_eye	left_eye, nose, right_ear
3	left_ear	left_eye, left_shoulder
4	right_ear	right_eye, right_shoulder
5	left_shoulder	left_hip, right_shoulder, left_elbow, left_ear
6	right_shoulder	right_hip, left_shoulder, right_elbow, right_ear
7	left_elbow	left_shoulder, left_wrist
8	right_elbow	right_shoulder, right_wrist
9	left_wrist	left_elbow
10	right_wrist	right_elbow
11	left_hip	left_knee, right_hip, left_shoulder
12	right_hip	right_knee, left_hip, right_shoulder
13	left_knee	left_ankle, left_hip
14	right_knee	right_ankle, right_hip
15	left_ankle	left_knee
16	right_ankle	right_knee

License

Non-commercial use only. Annotations: CC BY 4.0, the COCO Consortium. Images: the COCO Consortium does not own their copyright; each image comes from Flickr under its own license, recorded per image in COCO's annotation files, and most of them (68.9% of train2017) are Creative Commons NonCommercial. Use of the images must abide by the Flickr Terms of Use and each image's license. Cite Lin et al., Microsoft COCO: Common Objects in Context (ECCV 2014).

Source and attribution

The images and annotations come from COCO 2017, created by COCO Consortium (Lin et al., ECCV 2014) and released under Creative Commons Attribution 4.0. We are grateful to the original authors. If you use this data, please cite the original source above.

Citation

@inproceedings{lin2014coco,
  title = {Microsoft {COCO}: Common Objects in Context},
  author = {Lin, Tsung-Yi and Maire, Michael and Belongie, Serge and Hays, James and Perona, Pietro and Ramanan, Deva and Doll{\'a}r, Piotr and Zitnick, C. Lawrence},
  booktitle = {Computer Vision - ECCV 2014},
  pages = {740--755},
  publisher = {Springer},
  year = {2014}
}

---

Published from Pictograph - annotate, train, and deploy from one API.