Wiring Android's CameraX to a Quantized On-Device OCR Model for Real-Time Document Intelligence

Deep dive into combining CameraX preview streams with a quantized CRAFT+CRNN text detection/recognition pipeline — covering NNAPI vs GPU delegate selection for the two-stage model, text region proposal batching to avoid per-character inference overhead, and the frame-drop strategy that keeps end-to-end latency under 18ms on mid-range devices without draining the battery