MPII Human Pose JSON#

Format specification#

The original MPII Human Pose Dataset is available here.

Supported annotation types:

  • Bbox

  • Points

Supported attributes:

  • center (a list with two coordinates of the center point of the object)

  • scale (float)

Import MPII Human Pose Dataset (JSON)#

A Datumaro project with an MPII Human Pose Dataset (JSON) source can be created in the following way:

datum project create
datum project import --format mpii_json <path/to/dataset>

It is also possible to import the dataset using Python API:

import datumaro as dm

mpii_dataset = dm.Dataset.import_from('<path/to/dataset>', 'mpii_json')

MPII Human Pose Dataset (JSON) directory should have the following structure:

dataset/
├── jnt_visible.npy # optional
├── mpii_annotations.json
├── mpii_headboxes.npy # optional
├── mpii_pos_gt.npy # optional
├── 000000001.jpg
├── 000000002.jpg
├── 000000003.jpg
└── ...

Export to other formats#

Datumaro can convert an MPII Human Pose Dataset (JSON) into any other format Datumaro supports. To get the expected result, convert the dataset to a format that supports bounding boxes or points.

There are several ways to convert an MPII Human Pose Dataset (JSON) to other dataset formats using CLI:

datum project create
datum project import -f mpii_json <path/to/dataset>
datum project export -f voc -o ./save_dir -- --save-media

or

datum convert -if mpii_json -i <path/to/dataset> \
    -f voc -o <output/dir> -- --save-media

Or, using Python API:

import datumaro as dm

dataset = dm.Dataset.import_from('<path/to/dataset>', 'mpii_json')
dataset.export('save_dir', 'voc')

Examples#

Examples of using this format from the code can be found in the format tests