Flutter NSFW

Originally forked from the same named repo in github. MacOS supported is added to the repo.

• 1- Download, tflite model and put it in the assets folder
• 2 - Add the path of the tflite model to pubspec.yaml
• 3 - Read the file using path_provider plugin
• 4 - Make a seperate class for accessing the NSFW detector as

class NSFWDetector {
  NSFWDetector(this.modelPath, this.enableLog, this.isOpenGPU, this.numThreads);

  final String modelPath;
  final bool enableLog;
  final bool isOpenGPU;
  final int numThreads;

  bool isInitialized = false;

  Future<dynamic> detectInPhoto(String photoPath) async {
    if (!isInitialized) {
      Directory appDocDir = await getApplicationDocumentsDirectory();
      String appDocPath = appDocDir.path;
      var file = File(appDocPath + "/nsfw.tflite");
      if (!file.existsSync()) {
        var data = await rootBundle.load("assets/nsfw.tflite");
        final buffer = data.buffer;
        await file.writeAsBytes(
            buffer.asUint8List(data.offsetInBytes, data.lengthInBytes));
      }
      await FlutterNsfw.initNsfw(
        file.path,
      );
      isInitialized = true;
    }

    return FlutterNsfw.getPhotoNSFWScore(photoPath);
  }

  Future<dynamic> detectVideo(
    String videoPath,
    double nsfwThreshold,
    int width,
    int height,
  ) async {
    if (!isInitialized) {
      Directory appDocDir = await getApplicationDocumentsDirectory();
      String appDocPath = appDocDir.path;
      var file = File(appDocPath + "/nsfw.tflite");
      if (!file.existsSync()) {
        var data = await rootBundle.load("assets/nsfw.tflite");
        final buffer = data.buffer;
        await file.writeAsBytes(
            buffer.asUint8List(data.offsetInBytes, data.lengthInBytes));
      }
      await FlutterNsfw.initNsfw(
        file.path,
      );
      isInitialized = true;
    }
    final result = await FlutterNsfw.detectNSFWVideo(
        videoPath: videoPath,
        nsfwThreshold: nsfwThreshold,
        frameWidth: width,
        frameHeight: height,
        durationPerFrame: 1000);
    if (result != null) {
      print('the result is true');
      return result as bool;
    } else {
      print('this result is false');
      return false;
    }
  }
}

• 5 - Initiate the instance of the class that you have previously made

  NSFWDetector _nsfwDetector =
      NSFWDetector('assets/nsfw.tflite', true, true, 2);

• 6 - Make helper method for detecting nsfw photo,

  Future<dynamic> detectNSFWImage(String photo) async {
    final nsfwStatus = await _nsfwDetector.detectInPhoto(photo);
    if (nsfwStatus > 0.80) {
      return true;
    } else {
      return false;
    }
  }

• 7 - Make helper mothod for detecting NSFW video,

  Future<dynamic> detectNSFWVideo(String video, int width, int height) async {
    final nsfwStatus =
        await _nsfwDetector.detectVideo(video, 0.70, width, height);
    return nsfwStatus ?? false;
  }

• If you find that the model is increasing your App Size you can also host your model Firebase ML kit and download it using FirebaseMLModelDownloader
• If you are running the example app on emulator so it might not work because of GPU constraints please use a real device especially when testing for Android.