Add MediaPipe Pose backend to MGT-python

[Copilot]

MediaPipe offers modern, CPU-friendly pose estimation with 33 landmarks and no manual model management. This PR wires it into MGT-python as a first-class pose() backend alongside the existing OpenPose/Caffe path.

Core changes

• _pose_estimator.py — Rewrites MediaPipePoseEstimator to use the mediapipe Tasks API (mp.tasks.vision.PoseLandmarker), required since mediapipe 0.10 removed the old mp.solutions.pose.Pose. Model files (lite/full/heavy .task, ~8–28 MB) are auto-downloaded on first use and cached in musicalgestures/models/. Drops deprecated static_image_mode parameter.

• _pose.py — Connects MediaPipe to the existing pose() workflow:

  • Routes model='mediapipe' to a new _pose_mediapipe() helper (frame-by-frame FFmpeg pipe → MediaPipePoseEstimator → overlay skeleton → write video)
  • Adds MEDIAPIPE_POSE_CONNECTIONS (35 pairs covering all 33 landmarks) for skeleton rendering (cyan lines + red circles)
  • Adds _save_pose_txt() for exporting 33-landmark data to CSV/TSV/TXT with human-readable headers

• tests/test_new_features.py — Adds TestMediaPipePoseIntegration (10 tests covering connections, data export, model URLs, routing); fixes test_mediapipe_no_mediapipe_raises for the new attribute name.

Usage

import musicalgestures as mg

mv = mg.MgVideo("dancer.avi")
mv.pose(model="mediapipe")          # auto-downloads model on first call
mv.pose(model="mediapipe", data_format=["csv", "tsv"])  # multi-format export

# Or use the estimator directly
from musicalgestures import MediaPipePoseEstimator
est = MediaPipePoseEstimator(model_complexity=0)   # 0=lite, 1=full, 2=heavy
result = est.predict_frame(bgr_frame)              # PoseEstimatorResult, shape (33, 3)

Warning

Firewall rules blocked me from connecting to one or more addresses (expand for details)

I tried to connect to the following addresses, but was blocked by firewall rules:

• https://storage.googleapis.com/mediapipe-models/pose_landmarker/pose_landmarker_full/float16/1/pose_landmarker_full.task
  • Triggering command: /usr/bin/python python -m pytest tests/test_new_features.py::TestPoseEstimator -v get --global t pull.rebase (http block)
  • Triggering command: /usr/bin/python python -m pytest tests/test_new_features.py -v --global ame git (http block)
  • Triggering command: /usr/bin/python python -m pytest tests/ -v unset --global k/_temp/copilot-developer-action-main/dist/gh-gpgsign/gh-gpgsign-linux-x86_64 credential.usernld (http block)
• https://storage.googleapis.com/mediapipe-models/pose_landmarker/pose_landmarker_lite/float16/1/pose_landmarker_lite.task
  • Triggering command: /usr/bin/curl curl -s -o /home/REDACTED/work/MGT-python/MGT-python/musicalgestures/models/pose_landmarker_lite.task REDACTED -w %{http_code} nfig/composer/vendor/bin/git r (http block)
  • Triggering command: /usr/bin/curl curl -I REDACTED (http block)

If you need me to access, download, or install something from one of these locations, you can either:

• Configure Actions setup steps to set up my environment, which run before the firewall is enabled
• Add the appropriate URLs or hosts to the custom allowlist in this repository's Copilot coding agent settings (admins only)