Double Counting Parameters

When a model takes multiple inputs (ex. a Siamese network) that all go through the same set of layers before the results are compared later on in the network, the model summary double-counts all of the shared parameters. For example, in the network below, if layerA has 10 parameters and layerB has 5, then it will report 25 parameters instead of 15.

x1 -> layerA --v
........................+-> layerB -> y
x2 -> layerA --^