The ``ImageFeaturizer`` relies on a ``ONNXModel`` to do the
featurization of the image(s).  One can set this model directly using the
onnxModel parameter, or use setModel to set a known model name that will
be pulled from the ONNXHub.  See docs for other options to set the underlying
model.

The ``ImageFeaturizer`` takes an input column of images (the type
returned by the ``ImageReader``), automatically resizes them to fit the
``ONNXModel``'s inputs using an ``ImageTransformer``, and feeds them through
a pre-trained ONNX model. Some of the ``ImageTransformer`` parameters are
exposed on the featurizer so you can customize them.

You may specify intermediate model nodes by either using the ``headless``
parameter to automatically choose the full model output (headless = false),
or the intermediate feature nodes as indicated by the OnnxHub manifest for that
model (headless = true).
