在Windows上将tensorflow模型转换为coreml模型会产生问题

Question

I'm having problems with a conversion of a custom tf model. The model looks like this:

and after conversion(without errors) it looks like this:

Where is the add operator?

Another issue is that the output of the conversion gave me this:

Core ML input(s):
[name: "x_placeholder__0"
type {
  multiArrayType {
    shape: 41
    dataType: DOUBLE
  }
}
]
Core ML output(s):
[name: "softmax_prediction__0"
type {
  multiArrayType {
    shape: 2
    dataType: DOUBLE
  }
}
]

But my model has only float values? (conversion did actually change it from float32 to float64)

Can someone answer my questions and maybe tell me what I'm doing wrong?

Thx

Answer 1

A MatMul followed by an Add is the same thing as the innerProduct operator.

Note that the innerProduct layer has a "bias". If you look at those bias values, you'll see that these are the same values as are used in your Add operator. So coremltools / tf-coreml has simply combined these two operations into a single layer.

The default datatype for MLMultiArray objects is DOUBLE. You can change this to FLOAT, but that may not necessarily be any faster. Here's how you would do this in Python:

import coremltools  
import sys  

def update_multiarray_to_float32(feature):  
    if feature.type.HasField('multiArrayType'):  
        import coremltools.proto.FeatureTypes_pb2 as _ft  
        feature.type.multiArrayType.dataType = _ft.ArrayFeatureType.FLOAT32  

if __name__ == "__main__":  
    if len(sys.argv) != 3:  
        print "USAGE: %s <input_model_path> <output_model_path>" % sys.argv[0]  
        sys.exit(1)  

    input_model_path = sys.argv[1]  
    output_model_path = sys.argv[2]  

    spec = coremltools.utils.load_spec(input_model_path)  

    for input_feature in spec.description.input:  
        update_multiarray_to_float32(input_feature)  

    for output_feature in spec.description.output:  
        update_multiarray_to_float32(output_feature)  

    coremltools.utils.save_spec(spec, output_model_path)  

Script courtesy of a friendly Apple employee (see https://forums.developer.apple.com/thread/84401 ).