vision_ai_animals 0.1.1

On-device animal detection, species/breed classification, and pure-Dart analytics for Flutter, built on the vision_ai engine.

0.1.1

• Bumped to require vision_ai ^0.4.0 (which adds the body-pose engine). No change to animal detection, analytics, or this package's API.

0.1.0

Initial release — on-device animal detection for Flutter, built on vision_ai ^0.3.0.

Detection & classification

• Animal detection — bounding box + COCO class label + score per animal, via the MediaPipe Object Detector (EfficientDet-Lite0, LIVE_STREAM).
• Breed / species classification (AnimalConfig.detectBreed) — MediaPipe Image Classifier (EfficientNet-Lite, ImageNet) on each classifiable detection.
• AnimalVision.create(...) ergonomic session wrapper (start / stop / results / updateAnimalConfig / switchCamera / dispose); AnimalModels.ensureLoaded() copies the bundled models into app storage and returns their paths.

Analytics (pure Dart, zero model weight)

• AnimalTracker — stable cross-frame tracking IDs (greedy IoU).
• AnimalCounter — per-class + total counts.
• AnimalPresenceDetector — enter/exit events.
• AnimalMotionTracker — velocity, direction, and ActivityLevel (still/active/running).
• AnimalProximityEstimator — far / near / approaching.
• ZoneAlertDetector — enter/leave events for rectangular zones.
• cocoAnimalLabels preset, animalKindOf() / AnimalGrouping (pet/wildlife/other).

Overlays

• AnimalCameraView, AnimalBoxPainter, AnimalLabel, and AnimalOverlayStyle (boxes, class/breed labels, confidence bar, zones).

Bring your own model

• AnimalVision.create(customDetectorPath:, customClassifierPath:) swaps in your own TFLite models (skips the bundled-asset copy); allowAllCategories reports every class a custom (non-COCO) model emits.

Full model-capability utilization

• Breed classifier honors breedScoreThreshold, breedAllowedCategories, breedDeniedCategories — drop low-confidence guesses and constrain a general (ImageNet) classifier to the labels you care about.
• breedClassifiableCategories now includes bird by default, so parrots/birds are classified out of the box.
• displayNamesLocale returns localized category/breed names; rotationDegrees supports BYO models whose input isn't pre-oriented.
• AnimalResult carries categoryIndex and a localized displayName.

Bundled models

• object_detector.tflite (EfficientDet-Lite0) and breed_classifier.tflite (EfficientNet-Lite). float32 variants; swap to int8 or your own to shrink.

Platforms

• Verified working on Android and on a physical iOS device (iPhone 16 Pro Max); broader iOS device coverage welcome.