Skip to content

Body Eye Sync

Body Eye Sync is a Python desktop application for extracting synchronized computer-vision measurements from video. It tracks people in a recording, can run face detection and body-pose detection on the tracked person boxes, and saves the results in analysis-friendly Parquet files.

The project has two entry points:

  • body-eye-sync launches the Qt desktop application.
  • body-eye-sync-cli runs a saved experiment folder without opening the GUI.

The GUI is the primary workflow for opening a video, tuning the pipeline, running steps interactively, previewing overlays, and saving an experiment. The CLI uses the same experiment format for reproducible or batch processing.

Pipeline

The current pipeline is centered on video inputs:

  1. Object tracking detects and tracks objects across frames. By default it tracks COCO class 0, which is person.
  2. Face detection optionally finds one face per tracked person box and stores the face box, confidence, landmarks, and selected embeddings.
  3. Body-pose detection optionally estimates COCO keypoints inside tracked person boxes.

Outputs are written per input under the experiment's outputs/ directory.