dartcv #

OpenCV Bindings for Dart Language.

Important

The minimum required dart sdk version is 3.10 (Flutter 3.38) that supports hooks (Native-Assets).

Note

WIP, contributions are welcome!

dartcv

Example #

example

Supported Platforms #

Platform	Supported	Tested	Prebuilt Binaries
Android	✅	✅	x86_64, arm64-v8a, armeabi-v7a
iOS	✅	✅	arm64, x64+arm64(Simulator)
Linux	✅	✅	x64
Windows	✅	✅	x64
macOS	✅	✅	x64, arm64

Status #

Core Modules #

module	Binding status	Test status	description
core	✅	✅	Core module
calib3d	✅	✅	Calib3D module
dnn	✅	✅	DNN module
features2d	✅	✅	Features2D module
gapi	❌	❌	GAPI module
highgui	✅	✅	HighGUI module
imgcodecs	✅	✅	ImageCodecs module
imgproc	✅	✅	ImageProc module
ml	❌	❌	ML module
objdetect	✅	✅	Object Detection module
photo	✅	✅	Photo module
stitching	☑️	☑️	Stitching module
svd	✅	✅	SVD module
video	✅	✅	Video module
videoio	✅	✅	VideoIO module

Contrib Modules #

module	Binding status	Test status	description
aruco	✅	✅	ArUco module
img_hash	✅	✅	Image hashing module
cuda	❌	❌
wechat_qrcode	✅	✅
bgsegm	❌	❌
superres	❌	❌
xfeatures2d	❌	❌
ximgproc	✅	✅
xobjdetect	✅	✅
xphoto	❌	❌
quality	✅	✅
freetype	✅	✅

❌ : not finished
☑️ : partially supported
✅ : finished
modules not in the above table are not considered, contributions are welcome
VideoIO and HighGUI modules dynamically linked FFMPEG, you should be careful with the license, this project takes no responsibility for the license.

Usage #

Pure Dart

import 'package:dartcv4/dartcv.dart' as cv;

void main() {
  final img = cv.imread("test/images/lenna.png", flags: cv.IMREAD_COLOR);
  final gray = cv.cvtColor(img, gray, cv.COLOR_BGR2GRAY);
  print("${img.rows}, ${img.cols}");

  cv.imwrite("test_cvtcolor.png", gray);
}

Asynchronous

import 'package:dartcv4/dartcv.dart' as cv;
import 'dart:async';
void main() async {
  final img = await cv.imreadAsync("test/images/lenna.png", flags: cv.IMREAD_COLOR);
  final gray = await cv.cvtColorAsync(img, cv.COLOR_BGR2GRAY);
  print("${img.rows}, ${img.cols}");

  await cv.imwriteAsync("test_cvtcolor.png", gray);
}

Flutter

see example

~~More examples are on the way...~~ see awesome-opencv_dart and share yours

Configure hooks options

dartcv4 now supports hooks options, you can configure it in pubspec.yaml

hooks:
  user_defines:
    dartcv4:
      # debug: true
      # treeshake: true # tree-shaking unused symbols, onyl works in AOT mode and Dart 3.13+.
      include_modules: # `core` is always included
        - imgcodecs
        - imgproc
        # ...
      exclude_modules:
        - videoio
        - dnn
        # ...
      # whether to build OpenCV with OpenCL support, per platform
      windows:
        use_opencl: false
      linux:
        use_opencl: true
      # which OpenCV to build against, see below
      # opencv_version: "4.12.0"
      # opencv_dir: /path/to/opencv/lib/cmake/opencv4

Choosing which OpenCV to build against #

By default dartcv builds the OpenCV version this package pins, which is the one its bindings are developed and tested against. Two options change that, for projects that need a specific OpenCV — usually because their results have to match another environment, such as a Python service processing the same images.

Warning

The pinned version is the only one dartcv is developed and tested against. OpenCV changes its API between releases, so a different version may fail to compile, fail to link, or build cleanly and then behave differently at runtime. Problems that come from a non-default OpenCV are not supported by opencv_dart: before reporting one, reproduce it with the pinned version, and if it only happens with yours, it is yours to carry.

Use these options when you have a reason to accept that — matching another environment's results, or a platform SDK you do not control — and pin the version you tested, rather than tracking whatever is newest.

opencv_version: build this upstream tag from source instead of the pinned one, for example "4.12.0". Must be 4.12 or newer; dartcv calls APIs that do not exist before then, and the build stops with a message saying so.
opencv_dir: use an OpenCV that is already built, given as the directory holding OpenCVConfig.cmake. Takes precedence over opencv_version, and can also be set per platform, since a cross-compiled OpenCV lives somewhere different for each target:

hooks:
  user_defines:
    dartcv4:
      include_modules:
        - ximgproc
      android:
        opencv_dir: /path/to/OpenCV-android-sdk/sdk/native/jni
      ios:
        opencv_dir: /path/to/opencv-ios/device-arm64/lib/cmake/opencv4

Neither option changes anything for projects that do not set them.

debug: enable debug mode, default is false, if enabled, all messages will be printed to stderr.
treeshake: enable linker dead-code elimination, default is false. When enabled, the native library is compiled with function-level sections and the linker garbage-collects code that is not reachable from the exported symbols.
use_opencl: whether to build OpenCV with OpenCL support, default is false for all platforms. It can be configured per platform under a platform key (windows, linux, macos, android, ios). OpenCL is always disabled on ios since iOS does not support it. Enabling OpenCL may accelerate some operations (e.g. dnn), but it is disabled by default because the OpenCV OpenCL runtime can cause a racy hang-at-exit during process teardown; enable it only if you need the acceleration.
deployment_target: minimum Apple platform version, read from the ios / macos sub-map, e.g. ios: {deployment_target: "15.0"}. It overrides the deployment target dartcv would otherwise use, which comes from the target's code configuration (the deployment target of the Flutter/Xcode app). Quote values with trailing zeros ("10.10"): YAML parses an unquoted 10.10 as the number 10.1.
generator: the CMake generator to use, per platform under a platform key (windows, linux, macos, android, ios), one of Ninja, Unix Makefiles, Xcode, Visual Studio 16 2019, Visual Studio 17 2022, Visual Studio 18 2026. Defaults are platform specific: Unix Makefiles on Linux, Xcode on macOS/iOS, Ninja on Android and CMake's own default on Windows.
valid modules:
- core: always included
- included by default:
  - imgproc
  - imgcodecs
- excluded by default:
  - calib3d
  - features2d
  - flann
  - dnn
  - freetype
  - highgui
  - video
  - videoio
  - objdetect
  - photo
  - stitching
  - aruco
  - img_hash
  - quality
  - wechat_qrcode
  - ximgproc
  - xobjdetect
Note: even a module is excluded, it's dart code is still available, but throws a symbol not found exception when called.
~~videoio and highgui will introduce FFMPEG dynamic libraries (except for ios, ffmpeg is not supported on ios for now).~~ FFMPEG is no longer available with dartcv >= 2.2.0