face_recognition_engine 1.1.0

face_recognition_engine: ^1.1.0 copied to clipboard

Minor rather than patch: the dependency and SDK floors below are breaking for apps on older Flutter or Android.

Dependencies #

• Upgraded to camera ^0.12.0, google_mlkit_face_detection ^0.15.1 and image ^4.9.2.
• Raised minimum requirements to match: Flutter 3.44 / Dart 3.12, Android minSdkVersion 24 (was 21, required by camera), and iOS deployment target 15.5 (required by google_mlkit_face_detection). The README now states these in a Requirements table.

Example #

• The example is now runnable: added android/ and ios/ projects with camera permission (CAMERA, NSCameraUsageDescription), minSdk 24, iOS deployment target 15.5, and the Gradle JVM-target block. Drop a MobileFaceNet .tflite into example/assets/ — the directory is declared as a whole, so no pubspec edit is needed — and flutter run. See example/README.md.
• Stopped tracking (and therefore publishing) .flutter-plugins-dependencies, a generated file that embeds absolute local paths.

Documented #

• Added an Android JVM target mismatch section to the README. tflite_flutter compiles Java at 11 and camera_android_camerax at 17, while their Kotlin tasks default to the toolchain, which AGP rejects — consuming apps need a subprojects block pinning both to 17, placed before Flutter's generated evaluationDependsOn(":app") line.

Fixed #

• iOS support for the bundled camera screens. EnrollmentScreen and DetectionScreen previously always requested and decoded NV21, an Android format; on iOS the BGRA8888 frames were misread and embeddings were garbage. They now stream NV21 on Android and BGRA8888 on iOS.
• Added FaceRecognitionUtil.bgra8888ToImage (honours row padding via bytesPerRow) and FaceRecognitionUtil.cameraImageToImage, which picks the decoder from the frame's raw pixel format and returns null for formats it cannot decode.
• FaceRecognizer.embedCameraImage now decodes NV21 and BGRA8888, and throws UnsupportedError on other formats instead of embedding mis-decoded pixels.
• Fixed the FaceRecognizer dartdoc sample, which called create() with no arguments — that throws, since a model must be supplied.
• Reworked the README: badges, a features table, an accurate screenshots section, documented defaults for RecognitionConfig and SpoofDetector, and a new section on how guided enrollment picks its poses.
• Fixed incorrect README guidance: EnrollmentResult.photoJpg (encoded JPEG bytes) is not interchangeable with FaceProfileStore.savePhoto (which takes a decoded img.Image) — both routes are now shown.

• Added a screenshots section to the README showing the enrollment, liveness and live-recognition screens.

• Reordered the pub.flutter-io.cn screenshots so the identified-result dialog comes first and is used as the package thumbnail.

• Fixed the enrollment status text running two prompts together while capturing, e.g. "Hold still — capturing Turn your head right". It now reads "Hold still — capturing the right profile".
• Added screenshots of the enrollment, liveness and live-recognition flows to the pub.flutter-io.cn listing.

First public release.

• FaceRecognitionKit.enroll(context, ...) — opens a guided multi-angle camera and returns the captured embeddings (+ a front-pose JPEG) as an EnrollmentResult.
• FaceRecognitionKit.detect(context, candidates: ...) — opens a live camera, runs the configured liveness / anti-spoofing checks, and returns the matched FaceProfile as a DetectionResult.
• EnrollmentScreen / DetectionScreen widgets for custom navigation.
• FaceRecognizer — headless engine: load a model, embed a frame, identify probes.
• FaceProfileStore — file-backed multi-angle enrollment storage.
• RecognitionConfig — immutable, JSON-serialisable thresholds for recognition and liveness / anti-spoofing; drives both flows.
• SpoofDetector — passive texture/CNN anti-spoofing (bring your own model).
• FaceRecognitionUtil — low-level NV21 decode, crop, embed and cosine-match primitives.
• No model is bundled — supply a MobileFaceNet .tflite (112×112 → 192-d) via modelAsset: or modelBytes:. See the README "Model & license" section.