Flutter DocLayout Kit #

A Flutter plugin for document layout detection using PP-DocLayout model. This plugin provides on-device AI-powered document analysis for iOS and Android.

Screenshots #

End-to-end document layout workflow #

1. Offline layout On-device layout inference (~1s per image), no network required.	2. Static detection Analyzes document structure from test images (titles, text, tables, images).
3. Form reconstruction Turns detected layout blocks into editable fields and HTML-friendly structure.	4. Camera capture Capture photos and preview layout boxes directly in the camera view.
5. Real-time scan Continuous layout detection on live camera frames for scanning workflows.

Demo Video #

https://github.com/user-attachments/assets/7c348a4a-675f-46d2-b8a2-f798082b5245

Features #

• On-device document layout detection (no internet required)
• Detects 23 document element types (text, images, tables, formulas, etc.)
• High-performance native implementation using ONNX Runtime
• Supports image file and raw bytes input
• Cross-platform: iOS and Android

Supported Document Elements #

Installation #

1. Add Dependency #

From pub.dev (recommended):

dependencies:
  flutter_doclayout_kit: ^1.0.0

Or from GitHub:

dependencies:
  flutter_doclayout_kit:
    git:
      url: https://github.com/robert008/flutter_doclayout_kit.git

2. Download AI Model #

Download the ONNX model from GitHub Releases:

Steps:

1. Create assets/ folder in your project root
2. Download the model file and place it in assets/
3. Register in pubspec.yaml:

flutter:
  assets:
    - assets/pp_doclayout_m.onnx

Your project structure should look like:

your_app/
  assets/
    pp_doclayout_m.onnx    <-- AI model here
  lib/
  pubspec.yaml

3. Platform Setup #

iOS: Run pod install in your iOS directory. The native library will be downloaded automatically.

Android: The native library will be downloaded automatically on first build.

Usage #

Basic Usage #

import 'package:flutter_doclayout_kit/flutter_doclayout_kit.dart';

// Initialize with model path
final modelPath = await getModelPath(); // Get from your assets
DocLayoutKit.init(modelPath);

// Detect from image file
final result = DocLayoutKit.detectFromFile(
  '/path/to/document.jpg',
  confThreshold: 0.5,
);

// Process results
if (result.isSuccess) {
  print('Found ${result.count} elements');
  print('Inference time: ${result.inferenceTimeMs}ms');

  for (final box in result.detections) {
    print('${box.className}: ${box.score.toStringAsFixed(2)}');
    print('  Position: (${box.x1}, ${box.y1}) - (${box.x2}, ${box.y2})');
  }
}

Detect from Camera/Memory #

// For camera frames or in-memory images
final result = DocLayoutKit.detectFromBytes(
  imageBytes,
  width: 1920,
  height: 1080,
  channels: 3, // RGB
  confThreshold: 0.5,
);

Filter Results #

// Filter by element type
final tables = result.filterByClass(DocLayoutClass.table);
final images = result.filterByClass(DocLayoutClass.image);

// Filter by confidence score
final highConfidence = result.filterByScore(0.8);

API Reference #

DocLayoutKit #

DetectionResult #

DetectionBox #

Requirements #

• Flutter >= 3.3.0
• Dart >= 3.0.0
• iOS >= 12.0
• Android minSdk >= 21

Run Example App #

git clone https://github.com/robert008/flutter_doclayout_kit.git
cd flutter_doclayout_kit/example
flutter run

The example app will automatically download the required model and native libraries.

Building from Source #

If you want to build the native libraries yourself, the C++ source code is available in the src/ directory.

License #

Apache License 2.0

See LICENSE for details.

Acknowledgments #

• PP-DocLayout - Document layout analysis model by PaddlePaddle
• ONNX Runtime - Cross-platform inference engine
• OpenCV - Image processing library