I'm using onnxruntime-android within a Kotlin app, but am having trouble with using the result from an inference.
I can create arrays in Kotlin to construct the model and run inference, but when I get the result back as in
val result = ortSession.run(inputs)
the model outputs the result as a 2D tensor with one element
val probArray = result[0].value -> (this is a float[][])
To get the float I have to index into the array
val probability = probArray[0][0]
however this generates a compile time error. I need a way to cast probability into something Kotlin can use such as an Array<Array<Float>> but when I try to run
val probArray = result[0].value as Array<Array<Float>>
I get a runtime error for the cast
java.lang.ClassCastException: float[][] cannot be cast to java.lang.Float[][]
I understand the error, but cannot figure out a way to simply access the value inside probArray.
Was able to get this working with
val probArray = result[0].value as Array<FloatArray>
val probability = probArray[0][0]