sensitive_content_detection #

On-device sensitive-content detection for Flutter. Images are processed locally with the bundled TensorFlow Lite model and are not uploaded to a server.

Features #

• Detect images on Android and iOS.
• Analyze a File or encoded image bytes.
• Reuse one initialized detector for multiple images.
• Configure the confidence threshold, image size, and inference threads.

Installation #

Add the package to pubspec.yaml:

dependencies:
	sensitive_content_detection: ^0.1.0

Then fetch dependencies:

flutter pub get

The TensorFlow Lite model is included in the package. No separate model download is required.

Basic Usage #

import 'dart:io';

import 'package:sensitive_content_detection/sensitive_content_detection.dart';

final detector = SensitiveContentDetector();

Future<void> checkImage(File imageFile) async {
	try {
		final result = await detector.detect(imageFile);

		if (result.isSensitive) {
			print('Sensitive image: ${result.confidence}');
		} else {
			print('Image is safe: ${result.confidence}');
		}

		print(result.toJson());
	} finally {
		await detector.dispose();
	}
}

For repeated checks, create the detector once and dispose it when the owning widget or service is destroyed:

final detector = SensitiveContentDetector();

Future<DetectionResult> analyze(File imageFile) {
	return detector.detect(imageFile);
}

Future<void> close() => detector.dispose();

Analyze Image Bytes #

Use detectBytes when the image is already loaded in memory:

import 'dart:typed_data';

final Uint8List imageBytes = await imageFile.readAsBytes();
final result = await detector.detectBytes(imageBytes);

The bytes must contain a decodable image, such as JPEG or PNG data.

Configuration #

const config = SensitiveDetectionConfig(
	threshold: 0.40,
	maxImageDimension: 1024,
	numThreads: 2,
);

final detector = SensitiveContentDetector(config: config);

Configuration options:

Detection Results #

DetectionResult provides:

• isSensitive: whether the image meets the configured threshold.
• isSafe: the inverse of isSensitive.
• primaryCategory: SensitiveCategory.safe, sensitiveContent, or unknown.
• confidence: confidence for the selected primary category.
• detections: raw model detections.
• processingTime: elapsed processing time.
• modelVersion: model version when available.

Flutter Example #

The repository contains a complete example in the example folder. It uses image_picker to select a gallery image and passes the selected file to SensitiveContentDetector.detect.

Run it with:

cd example
flutter pub get
flutter run

Development #

Run the package tests from the repository root:

flutter test
flutter analyze

License #

See LICENSE. The bundled TensorFlow Lite model must be redistributed only when its model license permits it.