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

demo_compressed.mp4

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

ID	Element Type	ID	Element Type
0	paragraph_title	12	footnote
1	image	13	header
2	text	14	algorithm
3	number	15	footer
4	abstract	16	seal
5	content	17	chart_title
6	figure_title	18	chart
7	formula	19	formula_number
8	table	20	header_image
9	table_title	21	footer_image
10	reference	22	aside_text
11	doc_title

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:

Model	Size	Description
pp_doclayout_m.onnx	22 MB	Medium model (faster)
pp_doclayout_l.onnx	123 MB	Large model (more accurate)

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

Method	Description
init(String modelPath)	Initialize with ONNX model path
detectFromFile(String path, {double confThreshold})	Detect from image file
detectFromBytes(Uint8List data, {int width, int height, int channels, double confThreshold})	Detect from raw bytes
isInitialized	Check if initialized
version	Get library version

DetectionResult

Property	Type	Description
detections	List<DetectionBox>	List of detected elements
count	int	Number of detections
inferenceTimeMs	int	Inference time in ms
imageWidth	int	Original image width
imageHeight	int	Original image height
isSuccess	bool	Check if successful
hasError	bool	Check if has error

DetectionBox

Property	Type	Description
x1, y1, x2, y2	double	Bounding box coordinates
score	double	Confidence score (0-1)
classId	int	Element class ID (0-22)
className	String	Element class name
layoutClass	DocLayoutClass?	Enum value

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