advanced_image_processing_toolkit

Releases

1. ▸▾0.2.019 May 2026

   release 0.2.0: production hardening, drop com.example, fix ML Kit, sl…

   release 0.2.0: production hardening, drop com.example, fix ML Kit, sl…

   Open source →

   Additional notes2 sources agree

   Breaking

   • Android package renamed from com.example.advanced_image_processing_toolkit to dev.godfreylebo.advanced_image_processing_toolkit. Consumers do not need to change their Dart imports; the change is only visible to Android toolchains.
   • ObjectRecognition.detectObjects now takes a file path (String) instead of Uint8List. For raw bytes, use the new detectObjectsFromBytes(bytes, InputImageMetadata) which requires explicit image metadata (ML Kit cannot infer width/height/stride from a buffer). The previous bytes-based call would throw at runtime with zero-valued metadata; this version fixes that.
   • ObjectRecognition.drawDetections has been removed. It was a stub that returned the original bytes unchanged. Draw bounding boxes in your own UI layer (e.g. CustomPainter).
   • AR plugins are no longer bundled. arkit_plugin and arcore_flutter_plugin were always documented as "optional" but in fact pulled into every consumer app. They have been removed from this package's dependencies. To use AR, add them to your own app's pubspec.yaml.

   Changed

   • Replaced the umbrella google_ml_kit dependency with the individual google_mlkit_* packages — significant install-size reduction for consumer apps. Bumped to the latest 0.11–0.15 series.
   • iOS minimum version raised to 12.0 (required by Google ML Kit).
   • Android: compileSdk → 34, AGP → 7.4.2, Kotlin → 1.9.0, declared namespace for AGP 7+.
   • dart:io access guarded behind conditional imports so the package compiles cleanly on the web.

   Fixed

   • Version numbers were inconsistent across pubspec.yaml, the in-library version constant, the iOS podspec, and API_REFERENCE.md. All now read 0.2.0.
   • ObjectRecognition.detectObjects no longer passes Size(0, 0) and bytesPerRow: 0 to InputImageMetadata — these would have thrown on any real device.
   • README documented detectFaces, recognizeText, detectPoses, detectSurfaces, and named-parameter filter signatures that did not exist in the source. The README now matches the actual API surface.
   • API_REFERENCE.md was stuck on v0.1.2 / v0.0.6 examples and showed non-static final imageFilters = ImageFilters() usage. Fully rewritten against the v0.2.0 API.
   • Native getPlatformVersion is now implemented on both Android and iOS; previously the method handler only had TODO stubs for applyGrayscale / applyBlur and getPlatformVersion silently returned notImplemented.
   • Removed the parallel duplicate platform-interface and method-channel files in lib/src/ that shadowed the canonical ones in lib/.