Quickstart

1. Add a task package#

Choose the narrowest package that contains your task. Its dependency on mp_core supplies the shared model and data types.

dependencies:
  mp_vision: ^0.1.0

During pre-release development, use a Git dependency or a local path until the first packages are published.

2. Provide a model#

Models are application data, not SDK configuration. A model can come from a file, bytes, or a URL. For production browser builds, pin remote bytes with a digest.

final model = ModelAsset.uri(
  Uri.parse('https://example.com/models/efficientdet.tflite'),
  sha256: 'the-expected-lowercase-sha256-digest',
);

Native applications should normally download a versioned model to application storage, verify it, and pass its file path. Do not put private model URLs or tokens in client source.

3. Create and use the task#

import 'package:mp_core/mp_core.dart';
import 'package:mp_vision/mp_vision.dart';

Future<List<Detection>> detect(MpImage image) async {
  final detector = await ObjectDetector.create(
    ObjectDetectorOptions(
      baseOptions: BaseOptions(modelAsset: model),
    ),
  );
  try {
    final result = await detector.detect(image);
    return result.detections;
  } finally {
    await detector.close();
  }
}

Create long-lived task instances near the owning feature boundary rather than once per frame. A task serializes native calls so handles are never entered concurrently.

4. Select the execution mode#

• image accepts independent images.
• video accepts monotonically increasing timestamps.
• liveStream produces ordered asynchronous results where the platform exposes safe callback ownership. See live streams for adapter behavior.

5. Test without a model#

Every public task accepts an injectable runtime. Tests can substitute an implementation and verify application behavior without a network, GPU, or native SDK.

6. Run a complete example#

The examples page lists a runnable program for every package, from a plain dart run script to a live camera app. They are built in continuous integration and executed on a device, so they are the fastest way to confirm that the SDK works in your environment:

dart run examples/cli/bin/vision_face_detection.dart

Continue with the package pages for the full API surface of each task family.