🧘‍♀️Sportforma

Sportforma Dataset: a comprehensive collection of data (powered by Kopikat) for training pose estimation Computer Vision models

The dataset focuses on capturing human poses in various indoor and outdoor environments, catering to common scenarios encountered in everyday life. It comprises a diverse range of carefully generated images, ensuring a wide representation of body positions, poses, and orientations.

The dataset provides images and JSON with annotation. This link lets you get a dataset sample with 25 poses (4 views / 100 images).

The dataset contains 98322 images with 4 to 5 camera positions for each pose. We also provide visualization code to apply annotation on images.

The annotation is simple. You'll get 3D keypoints for the body, face and hands, and camera positions.

{
    "version": 1.1,
    "coordinate_system": "OpenGL",
    "camera_intrinsics": [],
    "scenes": [
        {
            "scene_id": "003494",
            "image": "images/scene_003494_view_000.png",
            "camera_pose": [],
            "kp_3d_body": [],
            "kp_3d_face": [],
            "kp_3d_hands": []
        },
        {
            "scene_id": "003494",
            "image": "images/scene_003494_view_001.png",
            "camera_pose": [],
            "kp_3d_body": [],
            "kp_3d_face": [],
            "kp_3d_hands": []
        },
        {...},
        {...},
        {...}
    ]
}

You can easily download the dataset sample by this link to see if it works for you.

Contact us at [e-mail] to get complete Sportforma under a commercial license.