Convert Lambda layer to Reshape layer in global avg pooling

[doronHarnoy]

The purpose of this PR is to convert Keras Lambda layers in Keras specific layers when converting global avg pooling operation

We are converting here Lambda layer that expanded dimensions twice (convert shape from [x1,x2,..xn] to [x1,x2,..xn,1,1])
to Keras reshape layer that will keep the same result

[bourcierj]

Hi,
I noted that in with a TensorFlow version > 2.6 we can use keepdims=True in tf.keras.layers.GlobalAveragePooling2D to avoid the need for an extra Reshape layer. FWIW, here is a link to a fix in my fork.

[doronHarnoy]

Hi, I noted that in with a TensorFlow version > 2.6 we can use keepdims=True in tf.keras.layers.GlobalAveragePooling2D to avoid the need for an extra Reshape layer. FWIW, here is a link to a fix in my fork.

Great man! thanks