monomech 0.15.17

Single camera biomechanics library

monomech

monomech is a notebook-first Python library for single-camera biomechanics. It helps you move from video or marker data into inspectable pose results, OpenSim-ready TRC files, inverse kinematics, inverse dynamics, and analysis tables without hiding the intermediate steps.

Full documentation: chags1313.github.io/monomech

Why Use It

• Start from a normal video or an existing TRC file.
• Export readable CSV and OpenSim-compatible TRC files.
• Run pose estimation, marker cleanup, scaling, inverse kinematics, and inverse dynamics as separate inspectable steps.
• Create OpenSim external loads from measured force data, arrays, carried loads, or estimated ground reaction forces.
• Export IK-driven OpenSim animations to a single portable GLB file.
• Review GLB meshes, markers, external-force arrows, IK traces, and ID traces in a Three.js HTML viewer.
• Keep OpenSim preflight checks on by default so NaNs and isolated gaps are fixed before IK and ID runs.
• Import the base package without installing heavy optional video or OpenSim dependencies.

Install

python -m pip install monomech

Choose extras only when you need them:

Workflow	Install command
Video pose estimation	python -m pip install "monomech[pose]"
OpenSim Python bindings	python -m pip install "monomech[opensim]"
OpenSim animation export	python -m pip install "monomech[animation]"
Notebooks and plots	python -m pip install "monomech[notebook]"
Everything optional	python -m pip install "monomech[all]"

monomech supports Python 3.10 through 3.12.

Quick Start: Video To TRC

from pathlib import Path
import monomech as mm

video_path = Path("data/subject01.mp4")
output_dir = Path("outputs/subject01")
output_dir.mkdir(parents=True, exist_ok=True)

pose3d_global = mm.estimate_pose(video_path, root_centered=False, floored=True)
pose3d_global = mm.gap_fill(mm.smooth(pose3d_global))

pose3d_global.to_csv(output_dir / "subject01_global.csv")
pose3d_global.to_trc(output_dir / "subject01_global.trc")

Or run the common video export path in one call:

run = mm.video_to_trc(video_path, output_dir=output_dir)

print(run.csv_paths)
print(run.trc_path)

Notebook-First Workflow

The high-level API is designed to read like the analysis steps in a notebook:

import monomech as mm

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

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

pose.vis_2d(frame=100)
pose.vis_3d(frame=100)

scaled_model = mm.run_scaling(pose, model="pose", output_dir="outputs/curl/scale")
ik = mm.run_ik(scaled_model, output_dir="outputs/curl/ik")

dumbbell = mm.load(type="carried", body="hand_r", mass_kg=10.0)
grf = mm.estimate_grf(pose, body_mass_kg=82.0)
forces = mm.external_forces(loads=[dumbbell, *grf])

id_result = mm.run_id(
    ik=ik,
    external_forces=forces,
    output_dir="outputs/curl/id",
)

ik.plot()
id_result.plot()

animation = mm.animate(
    ik=ik,
    id=id_result,
    external_loads_path=id_result.metadata["external_loads_mot_path"],
    output_dir="outputs/curl/visualizer",
    mode="balanced",
)
animation.show()

Carried and constant loads created without start_time and end_time are applied over the full IK trial. Use explicit times only for loads that turn on and off during the movement.

For batch scripts, use the one-call aliases:

result = mm.video_pipeline(
    "data/curl.mp4",
    model_path=mm.get_builtin_osim_model("pose"),
    output_dir="outputs/curl",
)

result.display()

Full Pipeline: Video To Inverse Dynamics

import monomech as mm

result = mm.video_to_inverse_dynamics(
    "data/subject01.mp4",
    model_path="models/subject01_scaled.osim",
    output_dir="outputs/subject01",
    body_mass_kg=75.0,
)

print(result.trc_path)
print(result.ik.path)
print(result.id.path)
print(result.visualizer.html_path)
result.display()

If you already have marker data in a TRC file, start at OpenSim:

result = mm.trc_to_inverse_dynamics(
    "outputs/subject01/subject01.trc",
    model_path="models/subject01_scaled.osim",
    output_dir="outputs/subject01/opensim",
    external_forces=None,
)

print(result.ik.path)
print(result.id.path)

Export the IK and ID run to one portable animation file:

animation = mm.save_opensim_animation(
    osim_path="models/subject01_scaled.osim",
    mot_path=result.ik.path,
    id_path=result.id.path,
    out_glb_path="outputs/subject01/animation/subject01_ik_id.glb",
    stride=2,
    decimate_target_reduction=0.35,
)

print(animation.glb_path)

Create a fast notebook-friendly Three.js dashboard with OpenSim body proxies, marker fallback, force arrows, IK plots, and ID plots:

viewer = mm.animate(
    ik=result.ik,
    id=result.id,
    model="models/subject01_scaled.osim",
    output_dir="outputs/subject01/visualizer",
    external_loads_path=result.id.metadata["external_loads_mot_path"],
    render="fast",
)
viewer.show()

The fast dashboard is a standalone HTML file, so it works in notebooks, local browsers, and GitHub Pages without waiting for GLB export. It controls lightweight body proxy meshes from OpenSim body transforms. When you need full anatomical surface meshes, use render="glb" or call mm.save_opensim_glb(...); the dashboard will reference the sibling GLB by default. Pass embed_glb=True only when you need one self-contained HTML file. The online visualizer starts empty and includes an Upload GLB control, which lets a reader drag in their own exported model without sending the file anywhere.

For a carried object plus estimated ground-reaction forces:

dumbbell = mm.external.carried_load(
    mass_kg=12.5,
    applied_to_body="radius_r",
    point=(0.0, -0.2, 0.0),
    name="right_dumbbell",
)

result = mm.video_to_inverse_dynamics(
    "data/curl.mp4",
    model_path=mm.get_builtin_osim_model("pose"),
    output_dir="outputs/curl",
    body_mass_kg=82.0,
    external_forces=mm.external.with_estimated_grf(dumbbell),
)

The GitHub Pages site includes an online GLB visualizer for quick review: chags1313.github.io/monomech/visualizer/.

Geometry note: the default full-body geometry is packaged with monomech, so most users do not need to pass geom_dir. Pass geom_dir only when using a different OpenSim model or custom mesh folder. If your model came from a macOS-created zip, avoid the __MACOSX folder because it usually contains only tiny ._*.vtp metadata files, not usable meshes.

For measured force plates, build an external-load spec from your force table:

right_grf = mm.external.from_csv(
    "data/right_force_plate.csv",
    applied_to_body="calcn_r",
    force_columns=("Fx", "Fy", "Fz"),
    point_columns=("Px", "Py", "Pz"),
    torque_columns=("Mx", "My", "Mz"),
    time_column="time",
    name="right_grf",
)

OpenSim Reliability Defaults

OpenSim is strict about missing or non-finite values. The OpenSim helpers preflight inputs by default:

• TRC marker gaps are interpolated before scale and IK.
• IK coordinate NaNs are interpolated before inverse dynamics.
• External-load data is resampled to IK time and non-finite force values are filled with zero.
• Preflight reports and generated paths are stored in result metadata.

print(ik.metadata["preflight"])
print(id_result.metadata["coordinate_preflight"])

Example Notebooks

The examples/ folder includes ready-to-edit notebooks:

• video_to_trc_quickstart.ipynb for video to CSV/TRC.
• marker_trc_cleanup.ipynb for TRC inspection, gap filling, and smoothing.
• opensim_scale_ik_template.ipynb for OpenSim scale and IK setup.
• video_to_inverse_dynamics_pipeline.ipynb for video to IK/ID with external loads.
• run_video_smoke.py for repeatable command-line checks on a real video.

Documentation

• Getting started
• Example notebooks
• API reference
• Online GLB visualizer
• External loads and forces
• OpenSim scale, IK, and ID
• OpenSim animation export
• Full video-to-ID pipeline
• Outputs and files

Development

python -m pip install -e ".[dev]"
python -m pytest tests
mkdocs build --strict
python -m build

Run a real video smoke test:

python examples/run_video_smoke.py "path/to/video.mp4" --output-dir outputs/smoke

Add --opensim when OpenSim-compatible bindings are installed.

Publishing

GitHub Actions builds distributions on every push to main. PyPI publishing goes directly to PyPI through trusted publishing and is triggered by version tags such as:

git tag v0.15.17
git push origin v0.15.17

The publish workflow can also be run manually from GitHub Actions with the publish input set to true.

License

See LICENSE.