Python keras.layers.concatenate() Examples

def RNNModel(vocab_size, max_len, rnnConfig, model_type):
	embedding_size = rnnConfig['embedding_size']
	if model_type == 'inceptionv3':
		# InceptionV3 outputs a 2048 dimensional vector for each image, which we'll feed to RNN Model
		image_input = Input(shape=(2048,))
	elif model_type == 'vgg16':
		# VGG16 outputs a 4096 dimensional vector for each image, which we'll feed to RNN Model
		image_input = Input(shape=(4096,))
	image_model_1 = Dropout(rnnConfig['dropout'])(image_input)
	image_model = Dense(embedding_size, activation='relu')(image_model_1)

	caption_input = Input(shape=(max_len,))
	# mask_zero: We zero pad inputs to the same length, the zero mask ignores those inputs. E.g. it is an efficiency.
	caption_model_1 = Embedding(vocab_size, embedding_size, mask_zero=True)(caption_input)
	caption_model_2 = Dropout(rnnConfig['dropout'])(caption_model_1)
	caption_model = LSTM(rnnConfig['LSTM_units'])(caption_model_2)

	# Merging the models and creating a softmax classifier
	final_model_1 = concatenate([image_model, caption_model])
	final_model_2 = Dense(rnnConfig['dense_units'], activation='relu')(final_model_1)
	final_model = Dense(vocab_size, activation='softmax')(final_model_2)

	model = Model(inputs=[image_input, caption_input], outputs=final_model)
	model.compile(loss='categorical_crossentropy', optimizer='adam')
	return model 

def weather_l2(hidden_nums=100,l2=0.01): 
    input_img = Input(shape=(37,))
    hn = Dense(hidden_nums, activation='relu')(input_img)
    hn = Dense(hidden_nums, activation='relu',
               kernel_regularizer=regularizers.l2(l2))(hn)
    out_u = Dense(37, activation='sigmoid',                 
                  name='ae_part')(hn)
    out_sig = Dense(37, activation='linear', 
                    name='pred_part')(hn)
    out_both = concatenate([out_u, out_sig], axis=1, name = 'concatenate')

    #weather_model = Model(input_img, outputs=[out_ae, out_pred])
    mve_model = Model(input_img, outputs=[out_both])
    mve_model.compile(optimizer='adam', loss=mve_loss, loss_weights=[1.])
    
    return mve_model 

def fire_module(x, fire_id, squeeze=16, expand=64):
    s_id = 'fire' + str(fire_id) + '/'

    if K.image_data_format() == 'channels_first':
        channel_axis = 1
    else:
        channel_axis = 3
    
    x = Conv2D(squeeze, (1, 1), padding='valid', name=s_id + sq1x1)(x)
    x = Activation('relu', name=s_id + relu + sq1x1)(x)

    left = Conv2D(expand, (1, 1), padding='valid', name=s_id + exp1x1)(x)
    left = Activation('relu', name=s_id + relu + exp1x1)(left)

    right = Conv2D(expand, (3, 3), padding='same', name=s_id + exp3x3)(x)
    right = Activation('relu', name=s_id + relu + exp3x3)(right)

    x = concatenate([left, right], axis=channel_axis, name=s_id + 'concat')
    return x


# Original SqueezeNet from paper. 

def _pad(self, input):
        """
        pads the network output so y_pred and y_true have the same dimensions
        :param input: previous layer
        :return: layer, last dimensions padded for 4
        """

        #pad = K.placeholder( (None,self.config.ANCHORS, 4))


        #pad = np.zeros ((self.config.BATCH_SIZE,self.config.ANCHORS, 4))
        #return K.concatenate( [input, pad], axis=-1)


        padding = np.zeros((3,2))
        padding[2,1] = 4
        return tf.pad(input, padding ,"CONSTANT")



    #loss function to optimize 

def CapsuleNet_v2(n_capsule = 10, n_routings = 5, capsule_dim = 16,
     n_recurrent=100, dropout_rate=0.2, l2_penalty=0.0001):
    K.clear_session()

    inputs = Input(shape=(200,))
    x = Embedding(20000, 300,  trainable=True)(inputs)        
    x = SpatialDropout1D(dropout_rate)(x)
    x = Bidirectional(
        CuDNNGRU(n_recurrent, return_sequences=True,
                 kernel_regularizer=l2(l2_penalty),
                 recurrent_regularizer=l2(l2_penalty)))(x)
    x = PReLU()(x)
    x = Capsule(
        num_capsule=n_capsule, dim_capsule=capsule_dim,
        routings=n_routings, share_weights=True)(x)
    x = Flatten(name = 'concatenate')(x)
    x = Dropout(dropout_rate)(x)
#     fc = Dense(128, activation='sigmoid')(x)
    outputs = Dense(6, activation='softmax')(x)
    model = Model(inputs=inputs, outputs=outputs)
    model.compile(loss='categorical_crossentropy', optimizer='nadam', metrics=['accuracy'])
    return model 

def CapsuleNet(n_capsule = 10, n_routings = 5, capsule_dim = 16,
     n_recurrent=100, dropout_rate=0.2, l2_penalty=0.0001):
    K.clear_session()

    inputs = Input(shape=(170,))
    x = Embedding(21099, 300,  trainable=True)(inputs)        
    x = SpatialDropout1D(dropout_rate)(x)
    x = Bidirectional(
        CuDNNGRU(n_recurrent, return_sequences=True,
                 kernel_regularizer=l2(l2_penalty),
                 recurrent_regularizer=l2(l2_penalty)))(x)
    x = PReLU()(x)
    x = Capsule(
        num_capsule=n_capsule, dim_capsule=capsule_dim,
        routings=n_routings, share_weights=True)(x)
    x = Flatten(name = 'concatenate')(x)
    x = Dropout(dropout_rate)(x)
#     fc = Dense(128, activation='sigmoid')(x)
    outputs = Dense(6, activation='softmax')(x)
    model = Model(inputs=inputs, outputs=outputs)
    model.compile(loss='categorical_crossentropy', optimizer='nadam', metrics=['accuracy'])
    return model 

def network_unet(param):
    n_joints = param['n_joints']
    pose_dn = param['posemap_downsample']
    img_h = param['IMG_HEIGHT']
    img_w = param['IMG_WIDTH']

    src_in = Input(shape=(img_h, img_w, 3))
    pose_src = Input(shape=(img_h / pose_dn, img_w / pose_dn, n_joints))
    pose_tgt = Input(shape=(img_h / pose_dn, img_w / pose_dn, n_joints))

    x = unet(src_in, concatenate([pose_src, pose_tgt]), [64] + [128] * 3 + [256] * 7,
             [256, 256, 256, 128, 64])
    y = my_conv(x, 3, activation='tanh')

    model = Model(inputs=[src_in, pose_src, pose_tgt], outputs=[y])
    return model 

def preprocess(x):
    return K.concatenate([
        x[:,:,0:1] / 360.0,
        x[:,:,1:3],
        x[:,:,3:4] / 360.0, 
        x[:,:,4:6],
        x[:,:,6:18] / 360.0,
        x[:,:,18:19] - x[:,:,1:2],
        x[:,:,19:22],
        x[:,:,28:29] - x[:,:,1:2],
        x[:,:,29:30],
        x[:, :, 30:31] - x[:, :, 1:2],
        x[:, :, 31:32],
        x[:, :, 32:33] - x[:, :, 1:2],
        x[:, :, 33:34],
        x[:, :, 34:35] - x[:, :, 1:2],
        x[:, :, 35:41],
    ], axis=2) 

def test_tiny_concat_random(self):
        np.random.seed(1988)
        input_dim = 10
        num_channels = 6

        # Define a model
        input_tensor = Input(shape=(input_dim,))
        x1 = Dense(num_channels)(input_tensor)
        x2 = Dense(num_channels)(x1)
        x3 = Dense(num_channels)(x1)
        x4 = concatenate([x2, x3])
        x5 = Dense(num_channels)(x4)

        model = Model(inputs=[input_tensor], outputs=[x5])

        # Set some random weights
        model.set_weights([np.random.rand(*w.shape) for w in model.get_weights()])

        # Get the coreml model
        self._test_model(model) 

def test_tiny_concat_seq_random(self):
        np.random.seed(1988)
        max_features = 10
        embedding_dims = 4
        seq_len = 5
        num_channels = 6

        # Define a model
        input_tensor = Input(shape=(seq_len,))
        x1 = Embedding(max_features, embedding_dims)(input_tensor)
        x2 = Embedding(max_features, embedding_dims)(input_tensor)
        x3 = concatenate([x1, x2], axis=1)

        model = Model(inputs=[input_tensor], outputs=[x3])

        # Set some random weights
        model.set_weights([np.random.rand(*w.shape) for w in model.get_weights()])

        # Get the coreml model
        self._test_model(model, one_dim_seq_flags=[True]) 

def unet(x_in, pose_in, nf_enc, nf_dec):
    x0 = my_conv(x_in, nf_enc[0], ks=7)  # 256
    x1 = my_conv(x0, nf_enc[1], strides=2)  # 128
    x2 = concatenate([x1, pose_in])
    x3 = my_conv(x2, nf_enc[2])
    x4 = my_conv(x3, nf_enc[3], strides=2)  # 64
    x5 = my_conv(x4, nf_enc[4])
    x6 = my_conv(x5, nf_enc[5], strides=2)  # 32
    x7 = my_conv(x6, nf_enc[6])
    x8 = my_conv(x7, nf_enc[7], strides=2)  # 16
    x9 = my_conv(x8, nf_enc[8])
    x10 = my_conv(x9, nf_enc[9], strides=2)  # 8
    x = my_conv(x10, nf_enc[10])

    skips = [x9, x7, x5, x3, x0]
    filters = [nf_enc[10], nf_dec[0], nf_dec[1], nf_dec[2], nf_enc[3]]

    for i in range(5):
        out_sz = 8*(2**(i+1))
        x = Lambda(interp_upsampling, output_shape = (out_sz, out_sz, filters[i]))(x)
        x = concatenate([x, skips[i]])
        x = my_conv(x, nf_dec[i])

    return x 

def test_shared_vision(self):
        digit_input = Input(shape=(27, 27, 1))
        x = Conv2D(64, (3, 3))(digit_input)
        x = Conv2D(64, (3, 3))(x)
        out = Flatten()(x)

        vision_model = Model(inputs=[digit_input], outputs=[out])

        # then define the tell-digits-apart model
        digit_a = Input(shape=(27, 27, 1))
        digit_b = Input(shape=(27, 27, 1))

        # the vision model will be shared, weights and all
        out_a = vision_model(digit_a)
        out_b = vision_model(digit_b)

        concatenated = concatenate([out_a, out_b])
        out = Dense(1, activation="sigmoid")(concatenated)
        model = Model(inputs=[digit_a, digit_b], outputs=out)
        model.set_weights([np.random.rand(*w.shape) for w in model.get_weights()])
        self._test_model(model) 

def fire_module(x, fire_id, squeeze=16, expand=64):
    s_id = 'fire' + str(fire_id) + '/'

    if K.image_data_format() == 'channels_first':
        channel_axis = 1
    else:
        channel_axis = 3
    
    x = Convolution2D(squeeze, (1, 1), padding='valid', name=s_id + sq1x1)(x)
    x = Activation('relu', name=s_id + relu + sq1x1)(x)

    left = Convolution2D(expand, (1, 1), padding='valid', name=s_id + exp1x1)(x)
    left = Activation('relu', name=s_id + relu + exp1x1)(left)

    right = Convolution2D(expand, (3, 3), padding='same', name=s_id + exp3x3)(x)
    right = Activation('relu', name=s_id + relu + exp3x3)(right)

    x = concatenate([left, right], axis=channel_axis, name=s_id + 'concat')
    return x


# Original SqueezeNet from paper. 

def test_tiny_image_captioning(self):
        # use a conv layer as a image feature branch
        img_input_1 = Input(shape=(16, 16, 3))
        x = Conv2D(2, (3, 3))(img_input_1)
        x = Flatten()(x)
        img_model = Model(inputs=[img_input_1], outputs=[x])

        img_input = Input(shape=(16, 16, 3))
        x = img_model(img_input)
        x = Dense(8, name="cap_dense")(x)
        x = Reshape((1, 8), name="cap_reshape")(x)

        sentence_input = Input(shape=(5,))  # max_length = 5
        y = Embedding(8, 8, name="cap_embedding")(sentence_input)
        z = concatenate([x, y], axis=1, name="cap_merge")
        z = LSTM(4, return_sequences=True, name="cap_lstm")(z)
        z = TimeDistributed(Dense(8), name="cap_timedistributed")(z)

        combined_model = Model(inputs=[img_input, sentence_input], outputs=[z])
        self._test_model(combined_model, one_dim_seq_flags=[False, True]) 

def test_tiny_image_captioning_feature_merge(self):

        img_input_1 = Input(shape=(16, 16, 3))
        x = Conv2D(2, (3, 3))(img_input_1)
        x = Flatten()(x)
        img_model = Model([img_input_1], [x])

        img_input = Input(shape=(16, 16, 3))
        x = img_model(img_input)
        x = Dense(8, name="cap_dense")(x)
        x = Reshape((1, 8), name="cap_reshape")(x)

        sentence_input = Input(shape=(5,))  # max_length = 5
        y = Embedding(8, 8, name="cap_embedding")(sentence_input)
        z = concatenate([x, y], axis=1, name="cap_merge")

        combined_model = Model(inputs=[img_input, sentence_input], outputs=[z])
        self._test_model(combined_model, one_dim_seq_flags=[False, True]) 

def test_dense_elementwise_params(self):
        options = dict(modes=[add, multiply, concatenate, average, maximum])

        def build_model(mode):
            x1 = Input(shape=(3,))
            x2 = Input(shape=(3,))
            y1 = Dense(4)(x1)
            y2 = Dense(4)(x2)
            z = mode([y1, y2])
            model = Model([x1, x2], z)
            return mode, model

        product = itertools.product(*options.values())
        args = [build_model(p[0]) for p in product]
        print("Testing a total of %s cases. This could take a while" % len(args))
        for param, model in args:
            self._run_test(model, param) 

def build_discriminator(self):

        z = Input(shape=(self.latent_dim, ))
        img = Input(shape=self.img_shape)
        d_in = concatenate([z, Flatten()(img)])

        model = Dense(1024)(d_in)
        model = LeakyReLU(alpha=0.2)(model)
        model = Dropout(0.5)(model)
        model = Dense(1024)(model)
        model = LeakyReLU(alpha=0.2)(model)
        model = Dropout(0.5)(model)
        model = Dense(1024)(model)
        model = LeakyReLU(alpha=0.2)(model)
        model = Dropout(0.5)(model)
        validity = Dense(1, activation="sigmoid")(model)

        return Model([z, img], validity) 

def create_model():
    inputs = Input(shape=(length,), dtype='int32', name='inputs')
    embedding_1 = Embedding(len(vocab), EMBED_DIM, input_length=length, mask_zero=True)(inputs)
    bilstm = Bidirectional(LSTM(EMBED_DIM // 2, return_sequences=True))(embedding_1)
    bilstm_dropout = Dropout(DROPOUT_RATE)(bilstm)
    embedding_2 = Embedding(len(vocab), EMBED_DIM, input_length=length)(inputs)
    con = Conv1D(filters=FILTERS, kernel_size=2 * HALF_WIN_SIZE + 1, padding='same')(embedding_2)
    con_d = Dropout(DROPOUT_RATE)(con)
    dense_con = TimeDistributed(Dense(DENSE_DIM))(con_d)
    rnn_cnn = concatenate([bilstm_dropout, dense_con], axis=2)
    dense = TimeDistributed(Dense(len(chunk_tags)))(rnn_cnn)
    crf = CRF(len(chunk_tags), sparse_target=True)
    crf_output = crf(dense)
    model = Model(input=[inputs], output=[crf_output])
    model.compile(loss=crf.loss_function, optimizer=Adam(), metrics=[crf.accuracy])
    return model 

def build_model():
    base_model = InceptionResNetV2(include_top=False, weights='imagenet', input_shape=(img_size, img_size, channel),
                                   pooling='avg')
    image_input = base_model.input
    x = base_model.layers[-1].output
    out = Dense(embedding_size)(x)
    image_embedder = Model(image_input, out)

    input_a = Input((img_size, img_size, channel), name='anchor')
    input_p = Input((img_size, img_size, channel), name='positive')
    input_n = Input((img_size, img_size, channel), name='negative')

    normalize = Lambda(lambda x: K.l2_normalize(x, axis=-1), name='normalize')

    x = image_embedder(input_a)
    output_a = normalize(x)
    x = image_embedder(input_p)
    output_p = normalize(x)
    x = image_embedder(input_n)
    output_n = normalize(x)

    merged_vector = concatenate([output_a, output_p, output_n], axis=-1)

    model = Model(inputs=[input_a, input_p, input_n],
                  outputs=merged_vector)
    return model 

def inception_module(x,params,concat_axis,padding='same',data_format=DATA_FORMAT,dilation_rate=(1,1),activation='relu',use_bias=True,kernel_initializer='glorot_uniform',bias_initializer='zeros',kernel_regularizer=None,bias_regularizer=None,activity_regularizer=None,kernel_constraint=None,bias_constraint=None,weight_decay=None):
    (branch1,branch2,branch3,branch4)=params
    if weight_decay:
        kernel_regularizer=regularizers.l2(weight_decay)
        bias_regularizer=regularizers.l2(weight_decay)
    else:
        kernel_regularizer=None
        bias_regularizer=None
    #1x1
    if branch1[1]>0:
        pathway1=Conv2D(filters=branch1[1],kernel_size=(1,1),strides=branch1[0],padding=padding,data_format=data_format,dilation_rate=dilation_rate,activation=activation,use_bias=use_bias,kernel_initializer=kernel_initializer,bias_initializer=bias_initializer,kernel_regularizer=kernel_regularizer,bias_regularizer=bias_regularizer,activity_regularizer=activity_regularizer,kernel_constraint=kernel_constraint,bias_constraint=bias_constraint)(x)
    #1x1->3x3
    pathway2=Conv2D(filters=branch2[0],kernel_size=(1,1),strides=1,padding=padding,data_format=data_format,dilation_rate=dilation_rate,activation=activation,use_bias=use_bias,kernel_initializer=kernel_initializer,bias_initializer=bias_initializer,kernel_regularizer=kernel_regularizer,bias_regularizer=bias_regularizer,activity_regularizer=activity_regularizer,kernel_constraint=kernel_constraint,bias_constraint=bias_constraint)(x)
    pathway2=Conv2D(filters=branch2[1],kernel_size=(3,3),strides=branch1[0],padding=padding,data_format=data_format,dilation_rate=dilation_rate,activation=activation,use_bias=use_bias,kernel_initializer=kernel_initializer,bias_initializer=bias_initializer,kernel_regularizer=kernel_regularizer,bias_regularizer=bias_regularizer,activity_regularizer=activity_regularizer,kernel_constraint=kernel_constraint,bias_constraint=bias_constraint)(pathway2)

    #1x1->5x5
    pathway3=Conv2D(filters=branch3[0],kernel_size=(1,1),strides=1,padding=padding,data_format=data_format,dilation_rate=dilation_rate,activation=activation,use_bias=use_bias,kernel_initializer=kernel_initializer,bias_initializer=bias_initializer,kernel_regularizer=kernel_regularizer,bias_regularizer=bias_regularizer,activity_regularizer=activity_regularizer,kernel_constraint=kernel_constraint,bias_constraint=bias_constraint)(x)
    pathway3=Conv2D(filters=branch3[1],kernel_size=(5,5),strides=branch1[0],padding=padding,data_format=data_format,dilation_rate=dilation_rate,activation=activation,use_bias=use_bias,kernel_initializer=kernel_initializer,bias_initializer=bias_initializer,kernel_regularizer=kernel_regularizer,bias_regularizer=bias_regularizer,activity_regularizer=activity_regularizer,kernel_constraint=kernel_constraint,bias_constraint=bias_constraint)(pathway3)

    #3x3->1x1
    pathway4=MaxPooling2D(pool_size=(3,3),strides=branch1[0],padding=padding,data_format=DATA_FORMAT)(x)
    if branch4[0]>0:
        pathway4=Conv2D(filters=branch4[0],kernel_size=(1,1),strides=1,padding=padding,data_format=data_format,dilation_rate=dilation_rate,activation=activation,use_bias=use_bias,kernel_initializer=kernel_initializer,bias_initializer=bias_initializer,kernel_regularizer=kernel_regularizer,bias_regularizer=bias_regularizer,activity_regularizer=activity_regularizer,kernel_constraint=kernel_constraint,bias_constraint=bias_constraint)(pathway4)
    if branch1[1]>0:
        return concatenate([pathway1,pathway2,pathway3,pathway4],axis=concat_axis)
    else:
        return concatenate([pathway2, pathway3, pathway4], axis=concat_axis) 

def inception_module(x,params,concat_axis,padding='same',data_format=DATA_FORMAT,dilation_rate=(1,1),activation='relu',use_bias=True,kernel_initializer='glorot_uniform',bias_initializer='zeros',kernel_regularizer=None,bias_regularizer=None,activity_regularizer=None,kernel_constraint=None,bias_constraint=None,weight_decay=None):
    (branch1,branch2,branch3,branch4)=params
    if weight_decay:
        kernel_regularizer=regularizers.l2(weight_decay)
        bias_regularizer=regularizers.l2(weight_decay)
    else:
        kernel_regularizer=None
        bias_regularizer=None
    #1x1
    if branch1[1]>0:
        pathway1=Conv2D(filters=branch1[1],kernel_size=(1,1),strides=branch1[0],padding=padding,data_format=data_format,dilation_rate=dilation_rate,activation=activation,use_bias=use_bias,kernel_initializer=kernel_initializer,bias_initializer=bias_initializer,kernel_regularizer=kernel_regularizer,bias_regularizer=bias_regularizer,activity_regularizer=activity_regularizer,kernel_constraint=kernel_constraint,bias_constraint=bias_constraint)(x)
    #1x1->3x3
    pathway2=Conv2D(filters=branch2[0],kernel_size=(1,1),strides=1,padding=padding,data_format=data_format,dilation_rate=dilation_rate,activation=activation,use_bias=use_bias,kernel_initializer=kernel_initializer,bias_initializer=bias_initializer,kernel_regularizer=kernel_regularizer,bias_regularizer=bias_regularizer,activity_regularizer=activity_regularizer,kernel_constraint=kernel_constraint,bias_constraint=bias_constraint)(x)
    pathway2=Conv2D(filters=branch2[1],kernel_size=(3,3),strides=branch1[0],padding=padding,data_format=data_format,dilation_rate=dilation_rate,activation=activation,use_bias=use_bias,kernel_initializer=kernel_initializer,bias_initializer=bias_initializer,kernel_regularizer=kernel_regularizer,bias_regularizer=bias_regularizer,activity_regularizer=activity_regularizer,kernel_constraint=kernel_constraint,bias_constraint=bias_constraint)(pathway2)

    #1x1->5x5
    pathway3=Conv2D(filters=branch3[0],kernel_size=(1,1),strides=1,padding=padding,data_format=data_format,dilation_rate=dilation_rate,activation=activation,use_bias=use_bias,kernel_initializer=kernel_initializer,bias_initializer=bias_initializer,kernel_regularizer=kernel_regularizer,bias_regularizer=bias_regularizer,activity_regularizer=activity_regularizer,kernel_constraint=kernel_constraint,bias_constraint=bias_constraint)(x)
    pathway3=Conv2D(filters=branch3[1],kernel_size=(5,5),strides=branch1[0],padding=padding,data_format=data_format,dilation_rate=dilation_rate,activation=activation,use_bias=use_bias,kernel_initializer=kernel_initializer,bias_initializer=bias_initializer,kernel_regularizer=kernel_regularizer,bias_regularizer=bias_regularizer,activity_regularizer=activity_regularizer,kernel_constraint=kernel_constraint,bias_constraint=bias_constraint)(pathway3)

    #3x3->1x1
    pathway4=MaxPooling2D(pool_size=(3,3),strides=branch1[0],padding=padding,data_format=DATA_FORMAT)(x)
    if branch4[0]>0:
        pathway4=Conv2D(filters=branch4[0],kernel_size=(1,1),strides=1,padding=padding,data_format=data_format,dilation_rate=dilation_rate,activation=activation,use_bias=use_bias,kernel_initializer=kernel_initializer,bias_initializer=bias_initializer,kernel_regularizer=kernel_regularizer,bias_regularizer=bias_regularizer,activity_regularizer=activity_regularizer,kernel_constraint=kernel_constraint,bias_constraint=bias_constraint)(pathway4)
    if branch1[1]>0:
        return concatenate([pathway1,pathway2,pathway3,pathway4],axis=concat_axis)
    else:
        return concatenate([pathway2, pathway3, pathway4], axis=concat_axis) 

def discriminator(param):
    img_h = param['IMG_HEIGHT']
    img_w = param['IMG_WIDTH']
    n_joints = param['n_joints']
    pose_dn = param['posemap_downsample']

    x_tgt = Input(shape=(img_h, img_w, 3))
    x_src_pose = Input(shape=(img_h / pose_dn, img_w / pose_dn, n_joints))
    x_tgt_pose = Input(shape=(img_h / pose_dn, img_w / pose_dn, n_joints))

    x = my_conv(x_tgt, 64, ks=5)
    x = MaxPooling2D()(x) # 128
    x = concatenate([x, x_src_pose, x_tgt_pose])
    x = my_conv(x, 128, ks=5)
    x = MaxPooling2D()(x) # 64
    x = my_conv(x, 256)
    x = MaxPooling2D()(x) # 32
    x = my_conv(x, 256)
    x = MaxPooling2D()(x) # 16
    x = my_conv(x, 256)
    x = MaxPooling2D()(x) # 8
    x = my_conv(x, 256)  # 8

    x = Flatten()(x)

    x = Dense(256, activation='relu')(x)
    x = Dense(256, activation='relu')(x)
    y = Dense(1, activation='sigmoid')(x)

    model = Model(inputs=[x_tgt, x_src_pose, x_tgt_pose], outputs=y, name='discriminator')
    return model 

def __call__(self, w):
        other_weights = K.cast(K.ones(K.shape(w))[:-1], K.floatx())
        last_weight = K.cast(K.equal(K.reshape(w[-1, :], (1, K.shape(w)[1])), 0.), K.floatx())
        appended = K.concatenate([other_weights, last_weight], axis=0)
        w *= appended
        return w 

def deconv_block(self, input, channels, kernel_size, skip):
        deconv1 = self.deconv(input, channels, kernel_size, 2)

        if skip is not None:
            concat1 = concatenate([deconv1, skip], 3)
        else:
            concat1 = deconv1

        iconv1 = self.conv(concat1, channels, kernel_size, 1)

        return iconv1 

def test_dangling_merge_right(self):

        x1 = Input(shape=(4,), name="input1")
        x2 = Input(shape=(5,), name="input2")
        y1 = Dense(6, name="dense")(x2)
        z = concatenate([y1, x1])
        model = Model(inputs=[x1, x2], outputs=[z])

        model.set_weights([np.random.rand(*w.shape) for w in model.get_weights()])

        self._test_model(model) 

def _layer_Conv(self):
        self.add_body(0, """
def convolution(weights_dict, name, input, group, conv_type, filters=None, **kwargs):
    if not conv_type.startswith('layer'):
        layer = keras.applications.mobilenet.DepthwiseConv2D(name=name, **kwargs)(input)
        return layer
    elif conv_type == 'layers.DepthwiseConv2D':
        layer = layers.DepthwiseConv2D(name=name, **kwargs)(input)
        return layer
    
    inp_filters = K.int_shape(input)[-1]
    inp_grouped_channels = int(inp_filters / group)
    out_grouped_channels = int(filters / group)
    group_list = []
    if group == 1:
        func = getattr(layers, conv_type.split('.')[-1])
        layer = func(name = name, filters = filters, **kwargs)(input)
        return layer
    weight_groups = list()
    if not weights_dict == None:
        w = np.array(weights_dict[name]['weights'])
        weight_groups = np.split(w, indices_or_sections=group, axis=-1)
    for c in range(group):
        x = layers.Lambda(lambda z: z[..., c * inp_grouped_channels:(c + 1) * inp_grouped_channels])(input)
        x = layers.Conv2D(name=name + "_" + str(c), filters=out_grouped_channels, **kwargs)(x)
        weights_dict[name + "_" + str(c)] = dict()
        weights_dict[name + "_" + str(c)]['weights'] = weight_groups[c]
        group_list.append(x)
    layer = layers.concatenate(group_list, axis = -1)
    if 'bias' in weights_dict[name]:
        b = K.variable(weights_dict[name]['bias'], name = name + "_bias")
        layer = layer + b
    return layer""") 

def test_dangling_merge_left(self):

        x1 = Input(shape=(4,), name="input1")
        x2 = Input(shape=(5,), name="input2")
        y1 = Dense(6, name="dense")(x2)
        z = concatenate([x1, y1])
        model = Model(inputs=[x1, x2], outputs=[z])

        model.set_weights([np.random.rand(*w.shape) for w in model.get_weights()])

        self._test_model(model) 

def emit_Concat(self, IR_node, in_scope=False):
        inputs = ', '.join('%s' % self.parent_variable_name(IR_node, s) for s in IR_node.in_edges)
        if in_scope:
            code = "{:<15} = K.concatenate([{}])".format(
                IR_node.variable_name,
                inputs)
        else:
            code = self._emit_merge(IR_node, "concatenate")
        return code 

def __call__(self, w):
        other_weights = K.cast(K.greater_equal(w, 0)[:-1], K.floatx())
        last_weight = K.cast(K.equal(K.reshape(w[-1, :], (1, K.shape(w)[1])), 0.), K.floatx())
        appended = K.concatenate([other_weights, last_weight], axis=0)
        w *= appended
        return w 

def call(self, inputs, **kwargs):
        if len(self._layers) == 1:
            return self._layers[0](inputs)

        filters = K.int_shape(inputs)[self._channel_axis]
        splits = self._split_channels(filters, self.groups)
        x_splits = tf.split(inputs, splits, self._channel_axis)
        x_outputs = [c(x) for x, c in zip(x_splits, self._layers)]
        x = layers.concatenate(x_outputs, axis=self._channel_axis)
        return x