Webcam-first coarse attention primitives for Linux gaze experiments
Find a file
joseph Rosenbaum bd048e74b9 Add calibration, HUD overlay, integrate HUD into CLI preview
- calibrate.py: interactive 9-target calibration with profile generation
  - Collects per-target samples, computes offsets/means/std
  - Saves/loads JSON profiles, supports --preview crosshair display
- hud.py: rich OpenCV diagnostic overlay
  - AOI grid, gaze dot with crosshair, confidence bar
  - Event ticker, face status, coords, FPS counter
  - Fully configurable HUDConfig
- run_session.py: preview mode now uses HUDRenderer instead of bare text
- Tests: 32 total (calibrate + HUD + session + x11 + primitives + capture)
2026-04-25 03:20:50 -05:00
src/webcam_gaze_poc Add calibration, HUD overlay, integrate HUD into CLI preview 2026-04-25 03:20:50 -05:00
tests Add calibration, HUD overlay, integrate HUD into CLI preview 2026-04-25 03:20:50 -05:00
.gitignore Add CLI runner, JSONL logging, replay tool, X11 context, live webcam fix 2026-04-25 03:05:38 -05:00
pyproject.toml feat: add attention event primitives 2026-04-24 23:42:10 -05:00
README.md feat: add detector-backed capture session loop 2026-04-24 23:50:23 -05:00

Webcam Gaze POC

Local-only Linux webcam gaze experiment.

Initial scope: coarse attention primitives plus a detector-neutral capture adapter and pure session runner. No persistence by default.

Current modules:

  • webcam_gaze_poc.primitives: AOIs, observations, attention events, engine state.
  • webcam_gaze_poc.capture: fake-detector-friendly adapter plus optional MediaPipe FaceLandmarker factory.
  • webcam_gaze_poc.session: frame-source/sampler/engine loop with an event sink.

Run tests:

python -m pytest

Real webcam capture still needs optional dependencies (opencv-python, mediapipe) and the local FaceLandmarker model file.