nicetoolbox.detectors.method_detectors.mmpose.mmpose_framework_2d

MMPose 2D framework — single consolidated detector class.

Previously this file held six subclasses (HRNetw48, VitPose, VitPoseHuge, RTMPoseLAIC, RTMPoseWHolebody, RTMPoseMPII) that differed only in their class-level components list and get_per_component_keypoint_mapping(). They are now a single MMPose2D class whose components are declared in TOML config and whose keypoint mapping logic dispatches on the requested components.

Classes

MMPose2D

The MMPose2D class is a method detector for 2D pose estimation using the MMPose framework.

class nicetoolbox.detectors.method_detectors.mmpose.mmpose_framework_2d.MMPose2D(io: SequenceIO, data: SequenceData, sequence_context: SequenceRuntimeConfig, algorithm_instance: str)[source]

The MMPose2D class is a method detector for 2D pose estimation using the MMPose framework.

Its post_inference method computes 3D points via multi-view stereo triangulation. components is set per-instance from the TOML config (e.g. body-only models declare components = [“body_joints”], wholebody models declare [“body_joints”, “hand_joints”, “face_landmarks”]).

Initialize base method detector with references.

get_per_component_keypoint_mapping(keypoints_indices)[source]

Build per-component keypoint index/description maps based on the components this instance was configured to produce.

  • body_joints: when components include hand_joints or face_landmarks we treat the model as wholebody and union body + foot keypoints; otherwise we use the body block only.

  • hand_joints / face_landmarks: straightforward 1:1 mapping.

post_inference()[source]

Post-inference processing for pose estimation components such as body_joints, hand_joints, and face_landmarks.

See module docstring for the full description of steps and outputs.