Skip to content

Sunhill666/dart_ncnn_yolov8

Repository files navigation

dart_ncnn_yolov8

A plugin use ncnn to drive yolov8.

Getting Started

This project is a starting point for a Flutter FFI plugin, a specialized package that includes native code directly invoked with Dart FFI.

Project structure

This template uses the following structure:

  • src: Contains the native source code, and a CmakeFile.txt file for building that source code into a dynamic library.

  • lib: Contains the Dart code that defines the API of the plugin, and which calls into the native code using dart:ffi.

  • platform folders (android, ios, windows, etc.): Contains the build files for building and bundling the native code library with the platform application.

How to use

For now, the example app only have Android app.

Get dependencies of plugin and example app

dart pub get
cd example
flutter pub get

Get NCNN Model

Use Ultralytics export NCNN model without any modification for model structure.

yolo export model=yolov8n format=ncnn

and put into example\assets\yolo, example\lib\providers\ncnn_yolo_options.dart used to custom yout own model.

Add NCNN & OpenCV-mobile library

Take Android for example:

  1. Download NCNN (download ncnn-xxxxxxx-android[-vulkan].zip).
  2. Download opencv-mobile (download opencv-mobile-x.x.x-android.zip).
  3. Extract Zip file to dart_ncnn_yolov8/android/src/main/cpp.
  4. Change corresponding platform library path in dart_ncnn_yolov8/src/CMakeLists.txt.

Use ffigen to generate ffi bindings (Optional)

In project root folder:

dart run ffigen --config ffigen.yaml

Use Freezed to generate code (Optional)

In project root folder:

dart run build_runner build

Building and bundling native code

The pubspec.yaml specifies FFI plugins as follows:

  plugin:
    platforms:
      some_platform:
        ffiPlugin: true

This configuration invokes the native build for the various target platforms and bundles the binaries in Flutter applications using these FFI plugins.

This can be combined with dartPluginClass, such as when FFI is used for the implementation of one platform in a federated plugin:

  plugin:
    implements: some_other_plugin
    platforms:
      some_platform:
        dartPluginClass: SomeClass
        ffiPlugin: true

A plugin can have both FFI and method channels:

  plugin:
    platforms:
      some_platform:
        pluginClass: SomeName
        ffiPlugin: true

The native build systems that are invoked by FFI (and method channel) plugins are:

  • For Android: Gradle, which invokes the Android NDK for native builds.
    • See the documentation in android/build.gradle.
  • For iOS and MacOS: Xcode, via CocoaPods.
    • See the documentation in ios/dart_ncnn_yolov8.podspec.
    • See the documentation in macos/dart_ncnn_yolov8.podspec.
  • For Linux and Windows: CMake.
    • See the documentation in linux/CMakeLists.txt.
    • See the documentation in windows/CMakeLists.txt.

Binding to native code

To use the native code, bindings in Dart are needed. To avoid writing these by hand, they are generated from the header file (src/dart_ncnn_yolov8.h) by package:ffigen. Regenerate the bindings by running flutter pub run ffigen --config ffigen.yaml.

Invoking native code

Very short-running native functions can be directly invoked from any isolate. For example, see sum in lib/dart_ncnn_yolov8.dart.

Longer-running functions should be invoked on a helper isolate to avoid dropping frames in Flutter applications. For example, see sumAsync in lib/dart_ncnn_yolov8.dart.

Flutter help

For help getting started with Flutter, view our online documentation, which offers tutorials, samples, guidance on mobile development, and a full API reference.

Reference

About

A flutter plugin use ncnn to drive yolov8

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published