Does not work for Siamese networks

[GMTSE]

Here is a piece of code that creates a siamese network:

input_a <- layer_input(shape=input_shape, name="left")
input_b <- layer_input(shape=input_shape, name="right")

processed_a <- input_a %>% base_network
processed_b <- input_b %>% base_network

end_network <- list(processed_a,processed_b) %>%
  layer_concatenate() %>%
  layer_dropout(0.1) %>%
  layer_dense(2, activation='softmax')

model <- keras_model(inputs=list(input_a, input_b), outputs=end_network)

Here is the (correct) output of printing the model:

15:18:27 R > model
Model
Model: "model_3"
________________________________________________________________________________________________________________________________________________________________________________________________
Layer (type)                                                   Output Shape                              Param #                Connected to                                                    
================================================================================================================================================================================================
left (InputLayer)                                              [(None, 1)]                               0                                                                                      
________________________________________________________________________________________________________________________________________________________________________________________________
right (InputLayer)                                             [(None, 1)]                               0                                                                                      
________________________________________________________________________________________________________________________________________________________________________________________________
model_2 (Model)                                                (None, 128)                               16768                  left[0][0]                                                      
                                                                                                                                right[0][0]                                                     
________________________________________________________________________________________________________________________________________________________________________________________________
concatenate (Concatenate)                                      (None, 256)                               0                      model_2[3][0]                                                   
                                                                                                                                model_2[4][0]                                                   
________________________________________________________________________________________________________________________________________________________________________________________________
dropout_3 (Dropout)                                            (None, 256)                               0                      concatenate[0][0]                                               
________________________________________________________________________________________________________________________________________________________________________________________________
dense_5 (Dense)                                                (None, 2)                                 514                    dropout_3[0][0]                                                 
================================================================================================================================================================================================
Total params: 17,282
Trainable params: 17,282
Non-trainable params: 0
________________________________________________________________________________________________________________________________________________________________________________________________

Here is the resulting graph with your tool:

Your tool missed the fact that input_right is linked to model_2.

Also, it would be great to be able to draw models recursively (e.g. see the internals of model_2 here)