YOLO exported TFLite model not compatible with GPU delegate in Flutter app

Question

I exported YOLOv8 (using ultralytics package) model to TFLite format and I'm trying to use it in my Flutter app. I'm using tflite_flutter package to do so and there is no issue with running it without GPU delegate, but when adding GPU delegates it breaks on model loading (ArgumentError (Invalid argument(s): Unable to create interpreter.)). No such problem with dummy TF 2.12. model running on GPU.

I'm exporting the model using command:

yolo export model=yolov8m-2023-04-25.pt format=tflite

And trying to load it by using:

final gpuDelegateV2 = tfl.GpuDelegateV2(
    options: tfl.GpuDelegateOptionsV2(
        isPrecisionLossAllowed: false,
        inferencePreference: tfl.TfLiteGpuInferenceUsage.fastSingleAnswer,
        inferencePriority1: tfl.TfLiteGpuInferencePriority.minLatency,
        inferencePriority2: tfl.TfLiteGpuInferencePriority.auto,
        inferencePriority3: tfl.TfLiteGpuInferencePriority.auto,
        maxDelegatePartitions: 1));

var interpreterOptions = tfl.InterpreterOptions()
  ..addDelegate(gpuDelegateV2);
final interpreter = await tfl.Interpreter.fromAsset('yolov8.tflite',
    options: interpreterOptions);

Also, the TF Analyzer shows, that the exported model should be compatible with GPU:

tf.lite.experimental.Analyzer.analyze(model_path="yolov8.tflite", gpu_compatibility=True)

I'm debugging it on Samsung Galaxy A52.

Does anyone know, what is wrong here?

YOLO exported TFLite model not compatible with GPU delegate in Flutter app

0 Answers0