flutter_onnxruntime #

Native Wrapper Flutter Plugin for ONNX Runtime

Current supported ONNX Runtime version: 1.22.0 (1.24.2 for iOS and macOS)

Note: For Android build, you need to upgrade your flutter_onnxruntime to version >=1.5.1 to satisfy the 16 KB Google Play compatibility requirement.

🌟 Why This Project? #

flutter_onnxruntime is a lightweight plugin that provides native wrappers for running ONNX Runtime on multiple platforms.

  📦 No Pre-built Libraries
  Libraries are fetched directly from official repositories during installation, ensuring they are always up-to-date!

  🛡️ Memory Safety
  All memory management is handled in native code, reducing the risk of memory leaks.

  🔄 Easy Upgrades
  Stay current with the latest ONNX Runtime releases without the hassle of maintaining complex generated FFI wrappers.

🚀 Getting Started #

Installation #

Add the following dependency to your pubspec.yaml:

dependencies:
  flutter_onnxruntime: ^1.6.0

Quick Start #

Example of running an addition model:

import 'package:flutter_onnxruntime/flutter_onnxruntime.dart';

// create inference session
final ort = OnnxRuntime();
final session = await ort.createSessionFromAsset('assets/models/addition_model.onnx');

// specify input with data and shape
final inputs = {
   'A': await OrtValue.fromList([1, 1, 1], [3]),
   'B': await OrtValue.fromList([2, 2, 2], [3])
}

// start the inference
final outputs = await session.run(inputs);

// print output data
print(await outputs['C']!.asList());

To get started with the Flutter ONNX Runtime plugin, see the API Usage Guide.

🧪 Examples #

Simple Addition Model #

A simple model with only one operator (Add) that takes two inputs and produces one output.

Run this example with:

cd example
flutter pub get
flutter run

Image Classification Model #

A more complex model that takes an image as input and classifies it into one of the predefined categories.

Clone this repository and run the example following the repo's guidelines.

📊 Component Overview #

🚧 Implementation Status #

✅: Completed

❌: Not supported

🚧: Ongoing

✍️: Planned

*: Retrieving model metadata and input/output info is not available for Swift and Javascript API.

¹: Execution Providers (EP) are hardware accelerated inference interface for AI inference (e.g., CPU, GPU, NPU, TPU, etc.)

📋 Required development setup #

Android #

Android build requires proguard-rules.pro inside your Android project at android/app/ with the following content:

-keep class ai.onnxruntime.** { *; }

or running the below command from your terminal:

echo "-keep class ai.onnxruntime.** { *; }" > android/app/proguard-rules.pro

Refer to troubleshooting.md for more information.

iOS #

ONNX Runtime requires minimum version iOS 16 and static linkage.

The plugin supports both CocoaPods (default) and Swift Package Manager. With Swift Package Manager enabled (see below), no Podfile changes are needed, but the app's iOS "Minimum Deployments" must be at least 16.0 (Xcode enforces the plugin's minimum platform under SPM).

For CocoaPods, in ios/Podfile, change the following lines:

platform :ios, '16.0'

# existing code ...

use_frameworks! :linkage => :static

# existing code ...

macOS #

macOS build requires minimum version macOS 14.

• For CocoaPods, in macos/Podfile, change the following lines:

  platform :osx, '14.0'
  

• Change the "Minimum Deployments" to 14.0 in XCode. In your terminal:

  open Runner.xcworkspace
  

  In Runner -> General, change Minimum Deployments to 14.0.

Swift Package Manager (iOS and macOS) #

The plugin supports Flutter's Swift Package Manager integration, which is opt-in on Flutter 3.24+:

flutter config --enable-swift-package-manager

Both install paths ship the same ONNX Runtime version and behave identically. CocoaPods remains fully supported and is used automatically when the flag is off or on older Flutter versions.

🛠️ Troubleshooting #

For troubleshooting, see the troubleshooting.md file.

🤝 Contributing #

Contributions to the Flutter ONNX Runtime plugin are welcome. Please see the CONTRIBUTING.md file for more information.

📚 Documentation #

Find more information in the documentation.