转换后 TFLite model 中缺少层

Question

So I trained a model using tensorflow and python and now I'm trying to use it in a C++ program. I use this code to convert my model to tflite (I don't have any error during the conversion):

model.load_weights('training_weights.h5', by_name=True)
model.save('saved_model/model')

converter = tf.lite.TFLiteConverter.from_saved_model("saved_model/model")
tflite_model = converter.convert()
with tf.io.gfile.GFile('model.tflite', 'wb') as f:
    f.write(tflite_model)

Then I load the model in C++ and I try to use it but the output doesn't match the output of my network. The last layers of the network, in python, are:

X = Lambda(lambda x: K.expand_dims(x, axis=2), name='deconv_expand_dim')(input_tensor)
X = Conv2DTranspose(filters, (kernel_size, 1), strides=(strides, 1), padding=padding, 
                    activation=activation, kernel_initializer=kernel_initializer,
                    bias_initializer=bias_initializer, name='deconv')(X)
X = Lambda(lambda x: K.squeeze(x, axis=2), name='deconv_reduce_dim')(X)

and the last layers of the model loaded, in C++, are:

1227 model/deconv_expand_dim/ExpandDims;StatefulPartitionedCall/model/deconv_expand_dim/ExpandDims
1228 model/deconv/Shape;StatefulPartitionedCall/model/deconv/Shape
1229 model/deconv/strided_slice;StatefulPartitionedCall/model/deconv/strided_slice1
1230 model/deconv/strided_slice_1;StatefulPartitionedCall/model/deconv/strided_slice_1
1231 model/deconv/strided_slice_2;StatefulPartitionedCall/model/deconv/strided_slice_22
1232 model/deconv/stack;StatefulPartitionedCall/model/deconv/stack
1233 model/deconv/conv2d_transpose;StatefulPartitionedCall/model/deconv/conv2d_transpose1
1234 model/deconv/BiasAdd;StatefulPartitionedCall/model/deconv/BiasAdd
1235 Identity

I just did a for loop to interpreter->tensors_size() to list the layers. Problems:

• interpreter->tensors_size() returns a greater number that varies for one test to another (around 1300).
• interpreter->outputs()[0] returns the index of Identity (1235).
• I have, in C++, the layers deconv_expand_dim and deconv but I don't have the layer deconv_reduce_dim .
• interpreter->tensor(outputIndex)->dims->size is equal to 0, which is problematic since this is the output of the network. Did I miss a step during the conversion? How do I get a valid output?

Any help is greatly appreciated, thanks.

Answer 1

I think I found the reason why it doesn't work. As you can see in the last layer I use the operator:

K.squeeze(x, axis=2)

which seems to not be supported yet with a specified axis ("tf.squeeze — As long as axis is not provided.", last-accessed:28 July 2020, link ).

That is why I don't have the same output in C++ as in Python, Tensorflow Lite doesn't support my network for the moment.