Conversion of tensorflow float32 is incompatible with a java object

Question

Error: Caused by: java.lang.IllegalArgumentException: Cannot convert between a TensorFlowLite tensor with type FLOAT32 and a Java object of type java.lang.String (which is compatible with the TensorFlowLite type STRING).

I have built a neural network from my dataset and have 2 layers then I have saved the model as h5 and then converted it into tflite using tf.keras model and conversions but when I deploy it in the application it gives me the above error

I have tried inputting with a lot of type of arrays and array lists

Error: Caused by: java.lang.IllegalArgumentException: Cannot convert between a TensorFlowLite tensor with type FLOAT32 and a Java object of type java.lang.String (which is compatible with the TensorFlowLite type STRING).

model.add(layers.Dense(500, input_dim=3, activation='relu'))
model.add(layers.Dense(1, activation= "relu"))
model.summary() #Print model Summary
model.compile(loss='mean_squared_error',optimizer='adam')
model.fit(X_train,Y_train,epochs=1000,validation_split=0.3)

How I convert:-

from tensorflow.contrib import lite
converter = lite.TFLiteConverter.from_keras_model_file( 'Model.h5')
tfmodel = converter.convert()
open ("model.tflite" , "wb") .write(tfmodel)

Implementation to android

ArrayList<String> list = new ArrayList<>();
list.add("-0.5698444");
list.add("-0.57369368");
list.add("-1.31490297");






try (Interpreter interpreter = new Interpreter(mappedByteBuffer)) {
    interpreter.run(list, "output");
}


private MappedByteBuffer loadModelFile() throws IOException {
    AssetFileDescriptor fileDescriptor = getAssets().openFd("model.tflite");
    FileInputStream inputStream = new FileInputStream(fileDescriptor.getFileDescriptor());
    FileChannel fileChannel = inputStream.getChannel();
    long startOffset = fileDescriptor.getStartOffset();
    long declaredLength = fileDescriptor.getDeclaredLength();
    return fileChannel.map(FileChannel.MapMode.READ_ONLY, startOffset, declaredLength);
}

It will be helpful if you post the code used to load the data in the TFLite model in Java. — Shubham Panchal
– Shubham Panchal, Commented May 27, 2019 at 13:06

Shubham Panchal · Accepted Answer · 2019-05-29 05:29:31Z

0

Found the mistake. In the line,

try (Interpreter interpreter = new Interpreter(mappedByteBuffer)) {
    interpreter.run(list, "output");
 }

The 2nd argument for the interpreter.run() has to be a float[] and not "output". The float[] will be populated with the class probabilities when the model is run.

The correct method to provide inputs and outputs to the interpreter.run() method:

Interpreter interpreter = new Interpreter(mappedByteBuffer)

float[][] inputs = new float[1][num_features]
// populate the inputs float array above
float[][] outputs = new float[1][num_classes]

interpreter.run( inputs , outputs )

float[] classProb = outputs[0]

answered May 29, 2019 at 5:29

Shubham Panchal

4,3092 gold badges14 silver badges42 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

DHRUV BATTA Over a year ago

Hi I tried using my shape of tensor it is saying is (0,1) while my java's is (1,1)

Collectives™ on Stack Overflow

Conversion of tensorflow float32 is incompatible with a java object

1 Answer 1

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related