On a related note to that discussed in #1053 , in the ScatteringKeras2D, its the case that the output_type is assigned to the backend variable, which isn't ultimately used by the ScatteringKeras2D object as it uses the ScatteringTensorFlow2D object to compute the actual scattering. We should fix this to use the argument names in the initialization call of the ScatteringBase2D object, just to make sure that we are passing in arguments correctly.

I also have to wonder if we need the Scattering Base, or if we could just subclass Tensorflow2D instead. I suspect that there was a reason we coded things this way.