I convert a tensorflow model to mlmodel and test it on iphone6. When I set MLPredictionOptions.usesCPUOnly to NO, the output is same as expected, it is same with tensorflow's output and tfcoreml's. But when I set it to YES, the output is quite different from usesCPUOnly. Could anyone give any suggestion? is it a bug of CoreML?
My mtmodel parameters:
Inputs: Image(Color 128x128)
Outputs: MultiArray(Double 1 x 128 x 128)
mtmodel size: 125KB
Thanks in advance.