End-to-end pipeline: CameraX ImageAnalysis → CRNN/EAST text detection model quantized to INT8 → GPU delegate inference → structured text output with bounding boxes, covering frame buffering strategy, model input normalization, output decoding with CTC beam search, and the memory layout decisions that determine whether you hit 45ms or 120ms on a mid-range device
title: "Real-Time OCR on Android: CameraX to TFLite in 45ms"
published: true
description: "Wire CameraX ImageAnalysis to a quantized CRNN/EAST pipeline on Android. The tricks that separate 45ms from 120ms — INT8, GPU delegate, and buffer layout."
tags: android, kotlin, mobile, architecture
canonical_url: https://mvpfactory.co/blog/real-time-ocr-android-camerax-tflite-45ms
You’ve built a machine learning model. It’s accurate, it’s lightweight, and it works perfectly on...
Build an AI-Powered Document Scanner and OCR Pipeline with Kotlin A document scanner can...
This is a submission for DEV's Summer Bug Smash: Clear the Lineup powered by Sentry. ...
CodeScanner and VisionCameraView put full-screen scanning and embedded live analysis above Codename One's on-device vision APIs,…
Deep dive into how Compose Multiplatform's Skia-based image decoding on iOS bypasses platform-native caching (NSCache, ImageIO…
The problem Neither the iOS Simulator nor the Android Emulator has real camera hardware....
