HumanSkeletonMixin

HumanSkeletonMixin is the pose-estimation primitive implemented in src/autopipeline/components/primitives/human_skeleton.py.

It is a narrow reusable helper for producing pose landmarks used by body-structure metrics.

Mixin

Signature

Class Signature

HumanSkeletonMixin(**kwargs)

Constructor

Constructor Parameters

Key	Default	Meaning
static_mode	True	Run MediaPipe Pose in static-image mode.
model_complexity	0	MediaPipe Pose model complexity.
min_detection_confidence	0.5	Detection confidence threshold.

The implementation always sets:

• enable_segmentation=False

Methods

Public Methods

Method	Purpose
get_skeleton_np(image)	Run pose estimation and return landmark coordinates.

Methods

Method Reference

get_skeleton_np(image: Image.Image)

Input

Parameter	Meaning
image	PIL image containing a person.

Output

Returns:

• np.ndarray of pose landmarks with rows [x, y, z, visibility]
• None if no pose landmarks are found

The coordinates are normalized MediaPipe pose coordinates, not pixel-space coordinates.

Config

Minimal Integration Example

init_config:
  static_mode: true
  model_complexity: 1
  min_detection_confidence: 0.5

Failure Mode

Failure Semantics

Soft-failure behavior:

• no pose landmarks -> returns None

Initialization or MediaPipe runtime failures are not wrapped in a custom recovery layer.

Extension

Extension Notes

• Keep general-purpose pose extraction here.
• Keep metric-specific body alignment or scoring above this layer.
• If you add a new body metric, document whether it expects normalized landmarks or pixel-space landmarks after post-processing.