How armit26/velvet works
VELVET is a browser-based, local computer-vision interface. A user connects a webcam; the app initializes MediaPipe, processes video frames, and presents detected objects with bounding boxes and confidence scores in the perception inspector. The sampled app excerpt shows camera setup, detector initialization, result formatting, overlays, and inspector rendering; the README documents the broader frame-to-result workflow. Scene exploration, diagnostics, and simulation are represented as interface affordances, not established runtime capabilities.
Main entry points: App.tsx Perception workspace · detector.ts Detector setup · main.tsx React entry
Components 9
Application interface
- React entry
src/main.tsx - Perception workspace
src/App.tsx - Detection inspector
src/App.tsx - Detection overlays
src/App.tsx
Vision processing
- Detector setup
src/vision/detector.ts
Browser and model
- Browser camera
- MediaPipe vision runtime
- Object detection model
Other components
- User
Connections 8
- React entry → Perception workspace: renders
src/main.tsx - User → Perception workspace: connects camera
README.md - Perception workspace → Browser camera: requests video
README.md - Perception workspace → Detector setup: initializes
src/App.tsx - Detector setup → MediaPipe vision runtime: loads runtime
src/vision/detector.ts - Detector setup → Object detection model: loads model
src/vision/detector.ts - Perception workspace → Detection inspector: presents results
src/App.tsx - Perception workspace → Detection overlays: draws boxes
src/App.tsx