How do I implement a tflite model from you in flutter?

Question

How do I implement a tflite model from you in flutter?

Kilian Schulz 0

I built an azure custom vision tflite model, and now have problems implementing it in flutter.

For one, chatgpt keeps telling me that I have to run a sigmoid function myself on the output.

My model can recognize a basketball and a basket hoop. Now in flutter, the output is a tensor of dimension 1,13,13,35. I use the tflite_flutter model but it is a bit trickier to unpack the output tensor.

How do i get the output bounding boxes and probabilities from the model?

We would very much appreciate any help possible, since tonight is the deadline to submit our project. Thanks a lot and best wishes

Kilian

YutongTie-MSFT 53,966 Reputation points Moderator

2023-07-03T21:40:42.7333333+00:00

@Kilian Schulz Thanks for reaching out to us. May I know which service you are working on and is there any document we can refer to reproduce your issue?

Kilian Schulz 0

Thank you Yutong for the quick reply. Here is the code where I process my image to work with your tflite model (i use the tflite_flutter_helper as well as the tflite_flutter package).

Please let me know when you see inconsistencies in the image preparation pipeline:

For more of our code, here is the necessary file link in our github, thanks a lot in advance for any help: https://github.com/zoematr/Hoopster/blob/main/lib/screens/recording_screen2.dart

Future<void> processCameraFrame(
      CameraImage image, tfl.Interpreter interpreter) async {
    try {
      img.Image imago = ImageUtils.convertYUV420ToImage(image);
      imago = ImageUtils.resizeImageTo32(imago);
      var tensorImage = TensorImage.fromImage(imago);
      tensorImage = ImageProcessorBuilder()
          .add(NormalizeOp(0, 255))
          .build()
          .process(tensorImage);

      var outputShape = interpreter.getOutputTensor(0).shape;
      var outputType = interpreter.getOutputTensor(0).type;
      var outputBuffer = TensorBuffer.createFixedSize(outputShape, outputType);
      print(interpreter.getInputTensor(0).shape);
      interpreter.run(tensorImage.buffer, {0: outputBuffer.getBuffer()});
      print('ran interpreter');
// This print statement never runs,I get the error "Bad state,failed precondition"

Kilian Schulz 0 Reputation points

2023-07-05T09:51:09.8366667+00:00

the service i am using is a tflite model exported from your azure custom vision platform

Your answer

YutongTie-MSFT 53,966 Reputation points Moderator

2023-07-03T21:40:42.7333333+00:00

@Kilian Schulz Thanks for reaching out to us. May I know which service you are working on and is there any document we can refer to reproduce your issue?
Kilian Schulz 0 Reputation points

2023-07-04T10:55:07.2966667+00:00

Thank you Yutong for the quick reply. Here is the code where I process my image to work with your tflite model (i use the tflite_flutter_helper as well as the tflite_flutter package).

Please let me know when you see inconsistencies in the image preparation pipeline:

For more of our code, here is the necessary file link in our github, thanks a lot in advance for any help: https://github.com/zoematr/Hoopster/blob/main/lib/screens/recording_screen2.dart

Future<void> processCameraFrame( CameraImage image, tfl.Interpreter interpreter) async { try { img.Image imago = ImageUtils.convertYUV420ToImage(image); imago = ImageUtils.resizeImageTo32(imago); var tensorImage = TensorImage.fromImage(imago); tensorImage = ImageProcessorBuilder() .add(NormalizeOp(0, 255)) .build() .process(tensorImage); var outputShape = interpreter.getOutputTensor(0).shape; var outputType = interpreter.getOutputTensor(0).type; var outputBuffer = TensorBuffer.createFixedSize(outputShape, outputType); print(interpreter.getInputTensor(0).shape); interpreter.run(tensorImage.buffer, {0: outputBuffer.getBuffer()}); print('ran interpreter'); // This print statement never runs,I get the error "Bad state,failed precondition"
Kilian Schulz 0 Reputation points

2023-07-05T09:51:09.8366667+00:00

the service i am using is a tflite model exported from your azure custom vision platform

Share via

How do I implement a tflite model from you in flutter?

Your answer