java.lang.AssertionError: "Does not support data type INT32" in Android Studio

Question

I am developing an application in Android Studio using a TensorFlow Lite model. When running the app, I encounter the following error:

java.lang.AssertionError: TensorFlow Lite does not support data type INT32

Below is the relevant part of my code:

// Prepare input tensor
val inputFeature0 = TensorBuffer.createFixedSize(inputShape, DataType.FLOAT32)
inputFeature0.loadArray(flatArray)

// Run inference
val outputs = model?.process(inputFeature0)
val rawOutputBuffer = outputs?.outputFeature0AsTensorBuffer

// Extract raw data as IntArray or FloatArray based on the data type
val outputArray = when (rawOutputBuffer?.dataType) {
    DataType.INT32 -> rawOutputBuffer.intArray // Directly access INT32 data
    DataType.FLOAT32 -> rawOutputBuffer.floatArray.map { it.toInt() }.toIntArray() // Convert FloatArray to IntArray
    else -> throw IllegalArgumentException("Unsupported output tensor data type: ${rawOutputBuffer?.dataType}")
}

1. The input tensor is of type FLOAT32, and the input data is loaded correctly using TensorBuffer.createFixedSize()and loadArray().

2. When processing the model's output tensor (outputFeature0AsTensorBuffer), I added checks to handle both FLOAT32and INT32 outputs.

3. Despite this, the app crashes with the error indicating that TensorFlow Lite does not support INT32.

What I have tried:

• Ensured that the input tensor uses FLOAT32.
• Verified that the TensorFlow Lite model is compatible with the FLOAT32 data type.
• Checked the output tensor data type and added handling for both FLOAT32 and INT32.

I expected the model inference to run without issues since I have handled both FLOAT32 and INT32 output cases.

Questions:

• Why does TensorFlow Lite throw this error even when FLOAT32 is used for the input tensor?
• Is the error related to how TensorFlow Lite internally manages data types like tensor dimensions or metadata?
• How can I resolve this error and ensure that TensorFlow Lite runs inference successfully?