This PR adds support for outputting multiple elements from the low-level TFImageTransformer. With this PR, one can then take TensorFlow's object detectors and run them at scale with Spark.
This adds a new output mode ("sql"): under this mode, the output of running tensorflow is added as extra columns to the resulting dataframe and is not interpreted as images or vectors.
This PR adds support for outputting multiple elements from the low-level TFImageTransformer. With this PR, one can then take TensorFlow's object detectors and run them at scale with Spark.
This adds a new output mode ("sql"): under this mode, the output of running tensorflow is added as extra columns to the resulting dataframe and is not interpreted as images or vectors.
Includes test.