Using fast neural-style transfer models in iOS

It turns out that we'll have no problem using the fst_frozen_quantized.pb model file, generated in step 7, in an iOS app built with the TensorFlow-experimental pod as shown in Chapter 2, Classifying Images with Transfer Learning, but the pre-trained multi-style model file from the TensorFlow Magenta project, which we'll use in a later section of this chapter, won't load with the TensorFlow pod (as of January 2018)—it'll throw the following error when trying to load the multi-style model file:

Could not create TensorFlow Graph: Invalid argument: No OpKernel was registered to support Op 'Mul' with these attrs. Registered devices: [CPU], Registered kernels:
  device='CPU'; T in [DT_FLOAT]
   [[Node: transformer/expand/conv1/mul_1 = Mul[T=DT_INT32](transformer/expand/conv1/mul_1/x, transformer/expand/conv1/strided_slice_1)]]

We talked about the reason and how to use manually-built TensorFlow libraries to fix this error in Chapter 3, Detecting Objects...