registerKerasImageUDF example not working - graph placeholder nChaneels was not given a corresponding dataframe name

I am able to create and train and evaluate a custom keras model based on InceptionV3 outside of spark. When using the model as udf (after registering it with registerKerasImageUDF) I am getting the following error:

py4j.protocol.Py4JJavaError: An error occurred while calling o96.selectExpr.
: java.lang.Exception: The graph placeholder nChannels was not given a corresponding dataframe field name as input:hints: Map()

It led me to test the examples in the README and i am getting the same error.

I am doing the following:

sparkSession = SparkSession.builder \
    .master("local[*]") \
    .appName("KerasModel") \
    .getOrCreate()
sparkSession.sparkContext.setLogLevel("WARN")

from keras.applications import InceptionV3
from sparkdl.udf.keras_image_model import registerKerasImageUDF

registerKerasImageUDF("inceptionV3_udf", InceptionV3(weights="imagenet"))

imagesDf = ImageSchema.readImages("/path/to/jpgs")
imagesDf.selectExpr("inceptionV3_udf(image) as predictions").show()

also tested with preprocessing:

sparkSession = SparkSession.builder \
    .master("local[*]") \
    .appName("KerasModel") \
    .getOrCreate()
sparkSession.sparkContext.setLogLevel("WARN")

from keras.applications import InceptionV3
from sparkdl.udf.keras_image_model import registerKerasImageUDF

def keras_load_img(fpath):
    from keras.preprocessing.image import load_img, img_to_array
    import numpy as np
    img = load_img(fpath, target_size=(299, 299))
    return img_to_array(img).astype(np.uint8)

registerKerasImageUDF("inceptionV3_udf", InceptionV3(weights="imagenet"), keras_load_img)

schema = StructType([
    StructField("uri", StringType()),
    StructField("label", IntegerType())
])

imagesDf = sparkSession.read.csv("/path/to/csv/with/image/uris",schema=schema)
imagesDf.selectExpr("inceptionV3_udf(uri) as predictions").show()

Both versions result in the following error:

  File "/opt/spark/python/lib/pyspark.zip/pyspark/sql/dataframe.py", line 1216, in selectExpr
  File "/opt/spark/python/lib/py4j-0.10.7-src.zip/py4j/java_gateway.py", line 1257, in __call__
  File "/opt/spark/python/lib/pyspark.zip/pyspark/sql/utils.py", line 63, in deco
  File "/opt/spark/python/lib/py4j-0.10.7-src.zip/py4j/protocol.py", line 328, in get_return_value
py4j.protocol.Py4JJavaError: An error occurred while calling o96.selectExpr.
: java.lang.Exception: The graph placeholder nChannels was not given a corresponding dataframe field name as input:hints: Map()
    at org.tensorframes.impl.SchemaTransforms$$anonfun$get$1.apply(DebugRowOps.scala:55)
    at org.tensorframes.impl.SchemaTransforms$$anonfun$get$1.apply(DebugRowOps.scala:55)
    at scala.Option.getOrElse(Option.scala:121)
    at org.tensorframes.impl.SchemaTransforms$class.get(DebugRowOps.scala:54)
    at org.tensorframes.impl.SchemaTransforms$.get(DebugRowOps.scala:275)

I am using spark version 2.3.1 tried with both version 1.1.0 and 1.0.0 of spark deep learning

OS ubuntu 16.04 Python 3.6.5

Any ideas?

databricks / spark-deep-learning

registerKerasImageUDF example not working - graph placeholder nChaneels was not given a corresponding dataframe name #146