How can change the models two have two inputa.

[MammadTavakoli]

I want design models based on some famous models that give two inputs. You designed more of them and I want change them to get two inputs same time. can you help me how can I change your created models to do that?

[MammadTavakoli]

I used this functions:

def rename_layers(model, prefix):
        for i, layer in enumerate(model.layers):
            # Use the private attribute _name to change the layer name
            layer.name = f"{prefix}_{layer.name}"
        return model

def create_model(model_name, learning_rate, loss_function, optimizer, class_num=3):
    if backend.image_data_format() == 'channels_last':
        bn_axis = 4
    else:
        bn_axis = 1

    inputs_ct = Input((None, None, None, 1))
    x = ZeroPadding3D(padding=(3, 3, 3), name='inputs_1_conv1_pad')(inputs_ct)
    x = Conv3D(64, (7, 7, 7),
                      strides=(2, 2, 2),
                      padding='valid',
                      kernel_initializer='he_normal',
                      name='inputs_1_conv1')(x)
    x = BatchNormalization(axis=bn_axis, name='inputs_1_bn_conv1')(x)
    x = Activation('relu')(x)
    x_ct = MaxPooling3D((3, 3, 3), strides=(2, 2, 2))(x)

    #
    inputs_pet = Input((None, None, None, 1))
    x = ZeroPadding3D(padding=(3, 3, 3), name='inputs_2_conv1_pad')(inputs_pet)
    x = Conv3D(64, (7, 7, 7),
                      strides=(2, 2, 2),
                      padding='valid',
                      kernel_initializer='he_normal',
                      name='inputs_2_conv1')(x)
    x = BatchNormalization(axis=bn_axis, name='inputs_2_bn_conv1')(x)
    x = Activation('relu')(x)
    x_pet = MaxPooling3D((3, 3, 3), strides=(2, 2, 2))(x)

    # x = add([x_1, x_2])
    x = concatenate ([x_ct, x_pet])
    input_tensor = Activation('relu')(x)
    ##########

    modelPoint, preprocess_input = Classifiers.get(model_name)

    model = modelPoint(include_top=True, input_tensor=input_tensor, classes=class_num, weights=None)

    #compile new model
    model.compile(optimizer=optimizer(learning_rate=learning_rate), 
                    loss=loss_function, 
                    metrics=['acc']) 
    return model

# model_name = 'resnet50'
# model = create_model(model_name)
# model.summary()

def create_model_2_model_padding(model_name, learning_rate, loss_function, optimizer, class_num=3):
    modelPoint, preprocess_input = Classifiers.get(model_name)

    inputs_ct = Input((None, None, None, 1))
    inputs_ct = ZeroPadding3D(padding=(3, 3, 3), name='inputs_ct_inputs_1_conv1_pad')(inputs_ct)

    inputs_pet = Input((None, None, None, 1))
    inputs_ct = ZeroPadding3D(padding=(3, 3, 3), name='inputs_pet_inputs_1_conv1_pad')(inputs_ct)

    model_ct = modelPoint(include_top=False, input_tensor=inputs_ct)
    model_ct = rename_layers(model_ct, 'model_ct')

    model_pet = modelPoint(include_top=False, input_tensor=inputs_pet)
    model_pet = rename_layers(model_pet, 'model_pet')

    combined = concatenate([model_ct.output, model_pet.output])

    # flattened = Flatten()(combined)
    fc1 = Dense(128, activation='relu', name="AddedDense1")(combined)
    output = Dense(class_num, activation='softmax', name="AddedDense2")(fc1)

    model = Model(inputs=[inputs_ct, inputs_pet], outputs=output)

    #compile new model
    model.compile(optimizer=optimizer(learning_rate=learning_rate), 
                    loss=loss_function, 
                    metrics=['acc'])
    return model

# model_name = 'resnet50'
# model = create_model_2_model_padding(model_name)
# model.summary()

def create_model_2_model(model_name, learning_rate, loss_function, optimizer, class_num=3):
    modelPoint, preprocess_input = Classifiers.get(model_name)

    inputs_ct = Input((None, None, None, 1))
    inputs_pet = Input((None, None, None, 1))

    model_ct = modelPoint(include_top=False, input_tensor=inputs_ct)
    model_ct = rename_layers(model_ct, 'model_ct')

    model_pet = modelPoint(include_top=False, input_tensor=inputs_pet)
    model_pet = rename_layers(model_pet, 'model_pet')

    combined = concatenate([model_ct.output, model_pet.output])

    # flattened = Flatten()(combined)
    fc1 =  Dense(128, activation='relu', name="AddedDense1")(combined)
    output =  Dense(class_num, activation='softmax', name="AddedDense2")(fc1)

    model =  Model(inputs=[inputs_ct, inputs_pet], outputs=output)

    #compile new model
    model.compile(optimizer=optimizer(learning_rate=learning_rate), 
                    loss=loss_function, 
                    metrics=['acc'])
    return model

# model_name = 'resnet50'
# model = create_model_2_model(model_name)
# model.summary()

def create_model_2_model_patch_size(model_name, patch_x, patch_y, patch_z, learning_rate, loss_function, optimizer, class_num=3):
    modelPoint, preprocess_input = Classifiers.get(model_name)

    inputs_ct = Input((patch_x, patch_y, patch_z, 1))
    inputs_pet = Input((patch_x, patch_y, patch_z, 1))

    model_ct = modelPoint(include_top=False, input_tensor=inputs_ct)
    model_ct = rename_layers(model_ct, 'model_ct')

    model_pet = modelPoint(include_top=False, input_tensor=inputs_pet)
    model_pet = rename_layers(model_pet, 'model_pet')

    combined = concatenate([model_ct.output, model_pet.output])

    flattened =  Flatten()(combined)
    fc1 =  Dense(128, activation='relu', name="AddedDense1")(flattened)
    output =  Dense(class_num, activation='softmax', name="AddedDense2")(fc1)

    model =  Model(inputs=[inputs_ct, inputs_pet], outputs=output)

    #compile new model
    model.compile(optimizer=optimizer(learning_rate=learning_rate), 
                    loss=loss_function, 
                    metrics=['acc'])
    return model

# patch_x = 20
# patch_y = 20
# patch_z = 60
# model_name = 'resnet50'
# model = create_model_2_model_size(model_name, patch_x, patch_y, patch_z)
# model.summary()