Unhandled Exception: Bad state: failed precondition about tflite_flutter_plugin

[matheper]

The PR is already closed but I want to share my experience in case someone has the same problem.

I faced the same issue and after a long time debugging I noticed my model expects an image type TfLiteType.float32.

If you load the image using the static method fromImage, it will create a TensorImage type TfLiteType.uint8.

// _inputImage type TfLiteType.uint8
_inputImage = TensorImage.fromImage(image);

You should create the TensorImage object and load your image using the loadImage:

// _inputImage type TfLiteType.float32
_inputImage = TensorImage(TfLiteType.float32); 
_inputImage.loadImage(image);

[xellDart]

The problem in my case was that it did not correctly declare the outputs of my models, each model has different shapes of tensors, I put a sample of my code at the end

TensorImage tensorImage = TensorImage.fromFile(file);
      tensorImage = _imageProcessor.process(tensorImage);

      TensorBuffer output0 = TensorBuffer.createFixedSize(
          _interpreter.getOutputTensor(0).shape,
          _interpreter.getOutputTensor(0).type);
      TensorBuffer output1 = TensorBuffer.createFixedSize(
          _interpreter.getOutputTensor(1).shape,
          _interpreter.getOutputTensor(1).type);

      Map<int, ByteBuffer> outputs = {0: output0.buffer, 1: output1.buffer};

      _interpreter.runForMultipleInputs([tensorImage.buffer], outputs);
      List<double> regression = output0.getDoubleList();
      List<double> classificators = output1.getDoubleList();

[am15h]

The list of integers with length 57344 (Uint8List) is the bytes representation of the float data.

There are multiple ways to get the data in float form. Ideally, you should be using runForMultipleInputs like this,

TensorBuffer output0 = TensorBuffer.createFixedSize(interpreter.getOutputTensor(0).shape, interpreter.getOutputTensor(0).type);
TensorBuffer output1 = TensorBuffer.createFixedSize(interpreter.getOutputTensor(1).shape, interpreter.getOutputTensor(1).type);

Map<int, ByteBuffer> outputs = {0: output0.buffer, 1: output1.buffer};

interpreter.runForMultipleInputs([tensorimage.buffer], outputs);

List<double> regression = output0.getDoubleList();

You can also create a TensorBufferFloat and then use loadBuffer(regressors.buffer, shape:interpreter.getOutputTensor(0).shape) and then use getDoubleList().

Hope this helps.

[am15h]

Hi, @KiruthikaAdhi.

TensorBuffer probabilityBuffer = TensorBuffer.createFixedSize([1, 1001], TfLiteType.uint8);

The type of this buffer should be TfLiteType.float32 and not TfLiteType.uint8.

More info:

This issue occurs because:

1. You might not be using the input/output buffers of correct shape, type as required by the interpreter. Verify this using interpreter.getInputTensors() and interpreter.getOutputTensors()

2. Probably the values of input buffer are not the expected ones. Maybe you are forgetting to normalize your input. Try using NormalizeOp.

[KiruthikaAdhi]

@xellDart @am15h Solved the issue by modifiying the shape of input and output tensors. Thanks for the help!

[am15h]

@xellDart This could be happening because of bad input to the interpreter. Are you sure that normalization is not required for the input image? Try adding NormalizeOp to the image processing pipeline with appropriate values. Have you tried to find out if the current values of tensorImage.buffer.asUint8List() are same as the expected ones?

[xellDart]

Thanks, already worked, I only have one more question...
I have this code in python:

regressors = interpreter.get_tensor(output_details[0]["index"])
classificators = interpreter.get_tensor(output_details[1]["index"])

And with your lib I can get this with

List<Tensor> tensors = _interpreter.getOutputTensors();
Tensor regressors = tensors[0];
Tensor classificators = tensors[1];

My question is, how i can get elements in tensor like this:
Python regressort tensor elements:

regressors: 
[ 3.4080482e-01  2.5098801e-02  2.8392452e+01 ... -5.0155647e+01
  4.7399345e+01  1.4660334e+01]

In dart i try with get data, but this return integers with length 57344, but when i get numElements, this returns correct value 14336
Thanks

[KiruthikaAdhi]

@xellDart @am15h

how did you solve this issue? I am facing the same error, my code is as follows

File imageFile = File(filePath);
ImageProcessor imageProcessor = ImageProcessorBuilder()
    .add(ResizeOp(224, 224, ResizeMethod.NEAREST_NEIGHBOUR))
    .build();
TensorImage tensorImage = TensorImage.fromFile(imageFile);
tensorImage = imageProcessor.process(tensorImage);
TensorImage outputImage = new TensorImage();
TensorBuffer probabilityBuffer = TensorBuffer.createFixedSize(<int>[1, 1001], TfLiteType.uint8);
Interpreter interpreter = await Interpreter.fromAsset("models/example.tflite");
interpreter.run(tensorImage.buffer, probabilityBuffer.buffer); File imageFile = File(filePath);

The error is as follows:

Unhandled Exception: Bad state: failed precondition
E/flutter ( 5476): #0 checkState (package:quiver/check.dart:71:5)
E/flutter ( 5476): #1 Tensor.setTo (package:tflite_flutter/src/tensor.dart:150:5)
E/flutter ( 5476): #2 Interpreter.runForMultipleInputs (package:tflite_flutter/src/interpreter.dart:194:33)
E/flutter ( 5476): #3 Interpreter.run (package:tflite_flutter/src/interpreter.dart:165:5)

I have edge detection model which takes an input image and gives an output image. How should I solve this?

[aklingam7]

Just look at the example repo here: https://github.com/am15h/tflite_flutter_helper/tree/master/example/image_classification

[CheerfulMinions]

@xellDart @am15h Solved the issue by modifiying the shape of input and output tensors. Thanks for the help!

So do I,
add

...
_inputImage = TensorImage(_inputType);
_inputImage.loadImage(image);

[rj76]

Hi all,

Sorry to still respond on this closed ticket, but I'm running into this while using TfLiteType.float32

The code that I'm trying is at:

https://github.com/TexMexMax/object_detection_flutter/blob/master/lib/tflite/classifierYolov4.dart#L246

print(_interpreter.getInputTensors());
print(_interpreter.getOutputTensors());

prints two times Tensor{_tensor: Pointer<TfLiteTensor>: address=0x72a8fa67c0, name: serving_default_input_1:0, type: TfLiteType.float32

Does anybody have an idea?

cheers

[cahyowhy]

this is my _interpreter.getInputTensors()

[Tensor{_tensor: Pointer: address=0xb4000072a21f3030, name: serving_default_images:0, type: uint8, shape: [1, 320, 320, 3], data: 307200}]

iam using both

_inputImage = TensorImage.fromImage(image);

_inputImage = TensorImage(TfLiteType.float32); 
_inputImage.loadImage(image);

and still got an issue 😭
need help