Data models

Defined in data_models/.

• Trajectory3d (trajectory_3d.py) is the currency of the posthoc pipeline: start_frame, end_frame, a (frames, markers, 3) triangulated_data array, plus reprojection_error and reprojection_error_by_camera arrays. from_observations() stacks per-frame observations and rejects gaps in frame numbers outright ("Observations are not contiguous"). save_to_arrays() writes {prefix}3d_data_spatial_xyz.npy, {prefix}reprojection_error.npy, and {prefix}reprojection_error_by_camera.npy. Within SkellyForge, only pipelines/dlc_pipeline.py calls it, to persist raw triangulation output. FreeMoCap's own triangulation step does not use it. It constructs Trajectory3d directly and saves results through the skellymodels Human manager instead.
• Observation3d holds one frame of named triangulated points with optional error data; to_point_dictionary() converts it to a dict of named Point3d values, skipping NaN markers.
• Trajectory2d is a thin (start, end, points) wrapper, and type_overloads.py aliases the dict shapes used to group 2D observations by camera and frame.