I have a custom object detection model.

I’m using Interpreter.runForMultipleInputsOutputs to detect an object on bitmap.

The problem is that at some point I started to receive zeroes at output when using GPU and
fine results with CPU.

An interesting fact that I have no problem with GPU on my test devece and I have this problem on remote device, that is my target device.

What could be the problem?

Here is my code, I have changed an exaple a bit:

var gpuDelegate: GpuDelegate? = if (device == Device.GPU) GpuDelegate() else null
val options = Interpreter
        .Options()
        .apply {
            when (device) {
                is Device.CPU -> {
                    setNumThreads(device.numThreads)
                }
                Device.GPU -> {
                    addDelegate(gpuDelegate)
                }
            }
        }

val interpreter: Interpreter = Interpreter(
        FileUtil.loadMappedFile(context, MODEL_FILENAME),
        options
    )

...

val inputTensor = processInputImage(bitmap, inputWidth, inputHeight)

val locations = arrayOf(Array(10) { FloatArray(4) })
val labelIndices = arrayOf(FloatArray(10))
val scores = arrayOf(FloatArray(10))
val outputCount = FloatArray(1)

val outputBuffer = mapOf(
    0 to locations,
    1 to labelIndices,
    2 to scores,
    3 to outputCount
)
val detectionRegions = mutableListOf<WeightRegion>()

interpreter.runForMultipleInputsOutputs(arrayOf(inputTensor.buffer), outputBuffer)

gpuDelegate?.close()
interpreter.close()

build.gradle:

android {
...
    buildFeatures {
        mlModelBinding true
    }
}

dependencies {
...
    implementation 'org.tensorflow:tensorflow-lite:2.7.0'
    implementation 'org.tensorflow:tensorflow-lite-gpu:2.7.0'
    implementation 'org.tensorflow:tensorflow-lite-support:0.3.0'
    implementation 'org.tensorflow:tensorflow-lite-metadata:0.3.0'
}