2D Pose¶

The usual entry point is mm.estimate_pose(). It estimates image-space 2D landmarks, keeps that result in the returned pose metadata, and uses it to build the global 3D pose.

2D pose overlay

import monomech as mm

pose = mm.estimate_pose(
    "data/subject01.mp4",
    root_centered=False,
    floored=True,
    sample_fps=30,
)

Source-Image Preview¶

pose.vis_2d() automatically uses the stored 2D pose and overlays the skeleton on the source video frame when the video path is available.

pose.vis_2d(frame=120)

The plot uses black landmarks and black segment lines, including a mid-hip to mid-shoulder trunk connection. The image coordinate system is preserved, so the preview matches the video frame instead of flipping vertically.

If you want to draw on a specific image array:

pose.vis_2d(frame=120, image=image_array)

Pass image=False to draw only the skeleton on a white background.

pose.vis_2d(frame=120, image=False)

Cleaning¶

The 2D and 3D result objects use the same cleaning API:

pose = mm.smooth(pose, cutoff_hz=6.0)
pose = mm.gap_fill(pose, max_gap_frames=12)

Use Pose2DConfig only when you need repeatable low-level pose-estimation settings:

config = mm.Pose2DConfig(stride=2, sample_fps=15)
pose = mm.estimate_pose("data/subject01.mp4", pose2d_config=config)

Outputs¶

pose.to_csv("outputs/subject01_global.csv")
pose.to_trc("outputs/subject01_global.trc")
display(pose.summary())