Internal shared infrastructure for the Vision Playground collection. NOT a public flagship.
This repo exists for one reason: to avoid rewriting the same infrastructure across the 10 flagship projects. It is public so the ecosystem's architecture is visible — but it is positioned as shared infrastructure, not a standalone library. Judge it as infrastructure.
Build abstractions from repetition, not from prophecy.
A component moves into this repo only when at least two projects genuinely need it. Until then, it lives inside the project that first needed it, written clean so it can be lifted later.
| Module | What it is |
|---|---|
camera |
webcam capture, mirror, resolution/FPS control |
trackers |
thin MediaPipe Tasks wrappers: Hand, Face (landmarks + blendshapes), Pose; swappable inference backend |
smoothing |
One-Euro / exponential filters for landmark streams |
perf |
FPS, frame time, inference time |
events |
event primitives: hysteresis, debounce, cooldown, confidence gates, stability windows |
mode |
the ~100-line Mode base class + folder-scan registry (pattern, not a framework) |
debug |
Experiment Mode overlay (telemetry strip, pipeline view, layer toggles, diagnostics stream, GIF export) |
Extracted on demand as projects prove the need (effects engine, media system, pack/config system, virtual camera, gesture engine, object tracker abstraction, audio analyzer). Premature abstractions — plugin frameworks, event buses, DI, PyPI publishing — are explicitly out of scope.
🚧 Scaffolding. Born from vision-filters (project 1); synced via vendored copy + sync script.