Error in converting keras model to core ml model.

Question

Created Jan ’21

Replies 0

Boosts 0

Participants 1

I have been trying to convert the Keras model to the core mlmodel. However, while doing so, I get multiple inputs and outputs. I just want the following as input and output.

Code Block name: "input"
type {  
	imageType {    
		width: 256    
		height: 64    
		colorSpace: GRAYSCALE  }
}
name: "output"
type {
  dictionaryType {
    stringKeyType {
    }
  }
}

The current model I get.

Code Block [name: "image"type {
  imageType {
    width: 256
    height: 64
    colorSpace: GRAYSCALE
  }
}
, name: "lstm1_h_in"
type {
  multiArrayType {
    shape: 256
    dataType: DOUBLE
  }
  isOptional: true
}
, name: "lstm1_c_in"
type {
  multiArrayType {
    shape: 256
    dataType: DOUBLE
  }
  isOptional: true
}
, name: "lstm1_h_in_rev"
type {
  multiArrayType {
    shape: 256
    dataType: DOUBLE
  }
  isOptional: true
}
, name: "lstm1_c_in_rev"
type {
  multiArrayType {
    shape: 256
    dataType: DOUBLE
  }
  isOptional: true
}
, name: "lstm2_h_in"
type {
  multiArrayType {
    shape: 256
    dataType: DOUBLE
  }
  isOptional: true
}
, name: "lstm2_c_in"
type {
  multiArrayType {
    shape: 256
    dataType: DOUBLE
  }
  isOptional: true
}
, name: "lstm2_h_in_rev"
type {
  multiArrayType {
    shape: 256
    dataType: DOUBLE
  }
  isOptional: true
}
, name: "lstm2_c_in_rev"
type {
  multiArrayType {
    shape: 256
    dataType: DOUBLE
  }
  isOptional: true
}
]
[name: "output"
type {
  dictionaryType {
    stringKeyType {
    }
  }
}
, name: "lstm1_h_out"
type {
  multiArrayType {
    shape: 256
    dataType: DOUBLE
  }
}
, name: "lstm1_c_out"
type {
  multiArrayType {
    shape: 256
    dataType: DOUBLE
  }
}
, name: "lstm1_h_out_rev"
type {
  multiArrayType {
    shape: 256
    dataType: DOUBLE
  }
}
, name: "lstm1_c_out_rev"
type {
  multiArrayType {
    shape: 256
    dataType: DOUBLE
  }
}
, name: "lstm2_h_out"
type {
  multiArrayType {
    shape: 256
    dataType: DOUBLE
  }
}
, name: "lstm2_c_out"
type {
  multiArrayType {
    shape: 256
    dataType: DOUBLE
  }
}
, name: "lstm2_h_out_rev"
type {
  multiArrayType {
    shape: 256
    dataType: DOUBLE
  }
}
, name: "lstm2_c_out_rev"
type {
  multiArrayType {
    shape: 256
    dataType: DOUBLE
  }
}
, name: "classLabel"
type {
  stringType {
  }
}
]

I am using this code to convert:

Code Block from keras.models import load_model
import coremltools as ct
class_labels = ["0", "1", "2", "3", "4", "5", "6", "7", "8", "9", "A", "B", "C", "D", "E", "F", "G", "H", "I", "J", "K",
                "L", "M", "N", "O", "P", "Q", "R", "S", "T", "U", "V", "W", "X", "Y", "Z", "a", "b", "c", "d", "e", "f",
                "g", "h", "i", "j", "k", "l", "m", "n", "o", "p", "q", "r", "s", "t", "u", "v", "w", "x", "y", "z", "-"]
tf_keras_model = load_model("HTRAdam.h5")
mlmodel = ct.converters.keras.convert('HTRAdam.h5', input_names=['image'], output_names=['output'],
                                                   class_labels=class_labels, image_input_names='image')
mlmodel.save('HTR.mlmodel')

When I enter the image in this, I don't get any output. I mean image output is nil.
I step into debug mode, but cannot find any error from model prediction.

I am guessing the issue might be with the compiling model but I have no idea about how to resolve it.
Thank you for your help.

Boost