Python keras.applications.VGG16 Examples

The following are 28 code examples of keras.applications.VGG16(). You can vote up the ones you like or vote down the ones you don't like, and go to the original project or source file by following the links above each example. You may also want to check out all available functions/classes of the module keras.applications , or try the search function .
Example #1
Source File: network.py    From Unified-Gesture-and-Fingertip-Detection with MIT License 6 votes vote down vote up
def model():
    model = VGG16(include_top=False, input_shape=(128, 128, 3))
    x = model.output

    y = x
    x = Flatten()(x)
    x = Dense(1024, activation='relu')(x)
    x = Dropout(0.5)(x)
    x = Dense(1024, activation='relu')(x)
    x = Dropout(0.5)(x)
    probability = Dense(5, activation='sigmoid', name='probabilistic_output')(x)

    y = UpSampling2D((3, 3))(y)
    y = Activation('relu')(y)
    y = Conv2D(1, (3, 3), activation='linear')(y)
    position = Reshape(target_shape=(10, 10), name='positional_output')(y)
    model = Model(input=model.input, outputs=[probability, position])
    return model 
Example #2
Source File: vgg.py    From inpainting-gmcnn-keras with MIT License 6 votes vote down vote up
def build_vgg_img_shape(y_pred, vgg_layers):
  input_shape = y_pred.shape.as_list()[1:4]
  img = Input(shape=input_shape)
  
  img_norm = _norm_inputs(img)
  vgg = VGG16(weights="imagenet", include_top=False)
  
  # Output the first three pooling layers
  vgg.outputs = [vgg.layers[i].output for i in vgg_layers]
  
  # Create model and compile
  model = Model(inputs=img, outputs=vgg(img_norm))
  model.trainable = False
  model.compile(loss='mse', optimizer='adam')
  
  return model 
Example #3
Source File: vgg.py    From inpainting-gmcnn-keras with MIT License 6 votes vote down vote up
def build_vgg_original_shape(y_pred, vgg_layers):
  input_shape = y_pred.shape.as_list()[1:4]
  img = Input(shape=input_shape)
  
  img_reshaped = Lambda(lambda x: tf.image.resize_nearest_neighbor(x, size=ORIGINAL_VGG_16_SHAPE))(
    img)
  
  img_norm = _norm_inputs(img_reshaped)
  vgg = VGG16(weights="imagenet", include_top=False)
  
  # Output the first three pooling layers
  vgg.outputs = [vgg.layers[i].output for i in vgg_layers]
  
  # Create model and compile
  model = Model(inputs=img, outputs=vgg(img_norm))
  model.trainable = False
  model.compile(loss='mse', optimizer='adam')
  
  return model 
Example #4
Source File: feature_detection.py    From deeposlandia with MIT License 6 votes vote down vote up
def vgg16(self):
        """Build the structure of a convolutional neural network from input
        image data to the last hidden layer on the model of a similar manner
        than VGG-net

        See: Simonyan & Zisserman, Very Deep Convolutional Networks for
        Large-Scale Image Recognition, arXiv technical report, 2014

        Returns
        -------
        tensor
            (batch_size, nb_labels)-shaped output predictions, that have to be
        compared with ground-truth values
        """
        vgg16_model = VGG16(input_tensor=self.X, include_top=False)
        y = self.flatten(vgg16_model.output, block_name="flatten")
        y = self.dense(y, 1024, block_name="fc1")
        y = self.dense(y, 1024, block_name="fc2")
        return self.output_layer(y, depth=self.nb_labels) 
Example #5
Source File: applications_test.py    From DeepLearning_Wavelet-LSTM with MIT License 5 votes vote down vote up
def test_vgg():
    app = random.choice([applications.VGG16, applications.VGG19])
    last_dim = 512
    _test_application_basic(app)
    _test_application_notop(app, last_dim)
    _test_application_variable_input_channels(app, last_dim)
    _test_app_pooling(app, last_dim) 
Example #6
Source File: file_util.py    From Mosaicer with MIT License 5 votes vote down vote up
def make_model(model, image_size):
    if model == "inceptionv3":
        base_model = InceptionV3(include_top=False, input_shape=image_size + (3,))
    elif model == "vgg16" or model is None:
        base_model = VGG16(include_top=False, input_shape=image_size + (3,))
    elif model == "mobilenet":
        base_model = MobileNet(include_top=False, input_shape=image_size + (3,))
    return base_model 
Example #7
Source File: solo_net.py    From Unified-Gesture-and-Fingertip-Detection with MIT License 5 votes vote down vote up
def model():
    model = VGG16(include_top=False, input_shape=(416, 416, 3))
    x = model.output
    x = Conv2D(1, (1, 1), activation='sigmoid')(x)
    output = Reshape((13, 13), name='output')(x)
    model = Model(model.input, output)
    return model 
Example #8
Source File: content_based_filtering.py    From keras-recommender with MIT License 5 votes vote down vote up
def __init__(self):
        self.matrix_res = None
        self.similarity_deep = None
        self.model = VGG16(include_top=False, weights='imagenet')
        self.matrix_idx_to_item_id = None
        self.item_id_to_matrix_idx = None 
Example #9
Source File: vgg.py    From inpainting-gmcnn-keras with MIT License 5 votes vote down vote up
def build_vgg16(y_pred, use_original_vgg_shape, vgg_layers):
  """
  Load pre-trained VGG16 from keras applications
  """
  if use_original_vgg_shape:
    return build_vgg_original_shape(y_pred, vgg_layers)
  else:
    return build_vgg_img_shape(y_pred, vgg_layers) 
Example #10
Source File: capgen.py    From Image-to-Image-Search with MIT License 5 votes vote down vote up
def __init__(self):
        self.checkpoint = pickle.load(open(CHECKPOINT_PATH, 'rb'),encoding='latin1')
        self.checkpoint_params = self.checkpoint['params']
        self.language_model = self.checkpoint['model']
        self.ixtoword = self.checkpoint['ixtoword']
        model = VGG16(weights="imagenet")
        self.visual_model = Model(input=model.input,output=model.layers[21].output)
        self.visual_model._make_predict_function()
        self.graph = tf.get_default_graph()
        self.BEAM_SIZE = 2 
Example #11
Source File: vgg16.py    From plaidbench with Apache License 2.0 5 votes vote down vote up
def build_model():
    import keras.applications as kapp
    from keras.layers import Input
    from keras.backend import floatx
    inputLayer = Input(shape=(224, 224, 3), dtype=floatx())
    return kapp.VGG16(input_tensor=inputLayer) 
Example #12
Source File: mlearn_for_image.py    From easy12306 with Artistic License 2.0 5 votes vote down vote up
def learn():
    (train_x, train_y, sample_weight), (test_x, test_y) = load_data()
    datagen = ImageDataGenerator(horizontal_flip=True,
                                 vertical_flip=True)
    train_generator = datagen.flow(train_x, train_y, sample_weight=sample_weight)
    base = VGG16(weights='imagenet', include_top=False, input_shape=(None, None, 3))
    for layer in base.layers[:-4]:
        layer.trainable = False
    model = models.Sequential([
        base,
        layers.BatchNormalization(),
        layers.Conv2D(64, (3, 3), activation='relu', padding='same'),
        layers.GlobalAveragePooling2D(),
        layers.BatchNormalization(),
        layers.Dense(64, activation='relu'),
        layers.BatchNormalization(),
        layers.Dropout(0.20),
        layers.Dense(80, activation='softmax')
    ])
    model.compile(optimizer=optimizers.RMSprop(lr=1e-5),
                  loss='sparse_categorical_crossentropy',
                  metrics=['accuracy'])
    model.summary()
    reduce_lr = ReduceLROnPlateau(verbose=1)
    model.fit_generator(train_generator, epochs=400,
                        steps_per_epoch=100,
                        validation_data=(test_x[:800], test_y[:800]),
                        callbacks=[reduce_lr])
    result = model.evaluate(test_x, test_y)
    print(result)
    model.save('12306.image.model.h5', include_optimizer=False) 
Example #13
Source File: applications_test.py    From DeepLearning_Wavelet-LSTM with MIT License 5 votes vote down vote up
def test_vgg():
    app = random.choice([applications.VGG16, applications.VGG19])
    last_dim = 512
    _test_application_basic(app)
    _test_application_notop(app, last_dim)
    _test_application_variable_input_channels(app, last_dim)
    _test_app_pooling(app, last_dim) 
Example #14
Source File: applications_test.py    From DeepLearning_Wavelet-LSTM with MIT License 5 votes vote down vote up
def test_vgg():
    app = random.choice([applications.VGG16, applications.VGG19])
    last_dim = 512
    _test_application_basic(app)
    _test_application_notop(app, last_dim)
    _test_application_variable_input_channels(app, last_dim)
    _test_app_pooling(app, last_dim) 
Example #15
Source File: applications_test.py    From DeepLearning_Wavelet-LSTM with MIT License 5 votes vote down vote up
def test_vgg():
    app = random.choice([applications.VGG16, applications.VGG19])
    last_dim = 512
    _test_application_basic(app)
    _test_application_notop(app, last_dim)
    _test_application_variable_input_channels(app, last_dim)
    _test_app_pooling(app, last_dim) 
Example #16
Source File: applications_test.py    From DeepLearning_Wavelet-LSTM with MIT License 5 votes vote down vote up
def test_vgg():
    app = random.choice([applications.VGG16, applications.VGG19])
    last_dim = 512
    _test_application_basic(app)
    _test_application_notop(app, last_dim)
    _test_application_variable_input_channels(app, last_dim)
    _test_app_pooling(app, last_dim) 
Example #17
Source File: applications_test.py    From DeepLearning_Wavelet-LSTM with MIT License 5 votes vote down vote up
def test_vgg():
    app = random.choice([applications.VGG16, applications.VGG19])
    last_dim = 512
    _test_application_basic(app)
    _test_application_notop(app, last_dim)
    _test_application_variable_input_channels(app, last_dim)
    _test_app_pooling(app, last_dim) 
Example #18
Source File: applications_test.py    From DeepLearning_Wavelet-LSTM with MIT License 5 votes vote down vote up
def test_vgg():
    app = random.choice([applications.VGG16, applications.VGG19])
    last_dim = 512
    _test_application_basic(app)
    _test_application_notop(app, last_dim)
    _test_application_variable_input_channels(app, last_dim)
    _test_app_pooling(app, last_dim) 
Example #19
Source File: generate_features.py    From show-attend-and-tell-keras with MIT License 5 votes vote down vote up
def setup_model(encoder, layer_name):
    image_input = Input(shape=(224, 224, 3))

    base_model = None
    if encoder == 'vgg16':
        base_model = VGG16(include_top=False, weights='imagenet', input_tensor=image_input, input_shape=(224, 224, 3))
    elif encoder == 'vgg19':
        base_model = VGG19(include_top=False, weights='imagenet', input_tensor=image_input, input_shape=(224, 224, 3))
    else:
        raise ValueError("not implemented encoder type")

    model = Model(inputs=base_model.input, outputs=base_model.get_layer(layer_name).output)
    return model 
Example #20
Source File: _validateSchema.py    From nyoka with Apache License 2.0 5 votes vote down vote up
def test_validate_keras_vgg(self):
        input_tensor = Input(shape=(224, 224, 3))
        model = VGG16(weights="imagenet", input_tensor=input_tensor)
        file_name = "keras"+model.name+".pmml"
        pmml_obj = KerasToPmml(model,dataSet="image",predictedClasses=[str(i) for i in range(1000)])
        pmml_obj.export(open(file_name,'w'),0)
        self.assertEqual(self.schema.is_valid(file_name), True) 
Example #21
Source File: train_multi.py    From DeepFashion with Apache License 2.0 5 votes vote down vote up
def save_bottlebeck_features_btl():

    dataset_btl_path = 'dataset_btl/train'
    batch_size = 1

    datagen = ImageDataGenerator(rescale=1. / 255)

    # build the VGG16 network
    model = applications.VGG16(include_top=False, weights='imagenet')                               # exclude 3 FC layers on top of network

    score_iou_btl_g, nb_btl_samples = get_images_count_recursive(dataset_btl_path)
    logging.debug('score_iou_btl_g {}'.format(score_iou_btl_g))
    logging.debug('nb_btl_samples {}'.format(nb_btl_samples))


    ## Train
    generator = datagen.flow_from_directory(
        dataset_btl_path,
        target_size=(img_width, img_height),
        batch_size=batch_size,
        classes=None,                                                                               #  the order of the classes, which will map to the label indices, will be alphanumeric
        class_mode=None,                                                                            # "categorical": 2D one-hot encoded labels; "None": yield batches of data, no labels; "sparse" will be 1D integer labels.
        save_to_dir='temp',
        shuffle=False)                                                                              # Don't shuffle else [class index = alphabetical folder order] logic used below might become wrong; first 1000 images will be cats, then 1000 dogs
    logging.info('generator.class_indices {}'.format(generator.class_indices))
                                                                                                    # classes: If not given, the order of the classes, which will map to the label indices, will be alphanumeric
    bottleneck_features_btl = model.predict_generator(
        generator, nb_btl_samples // batch_size)
    logging.debug('bottleneck_features_btl {}'.format(bottleneck_features_btl.shape))           # bottleneck_features_train (10534, 4, 4, 512) where train images i.e Blazer+Jeans=5408+5126=10532 images;

    # save the output as a Numpy array
    logging.debug('Saving bottleneck_features_btl...')
    np.save(open('output/bottleneck_features_btl.npy', 'w'),
            bottleneck_features_btl) 
Example #22
Source File: train_multi_v2.py    From DeepFashion with Apache License 2.0 5 votes vote down vote up
def save_bottlebeck_features_btl():

    dataset_btl_path = 'dataset_btl/train'
    batch_size = 1

    datagen = ImageDataGenerator(rescale=1. / 255)

    # build the VGG16 network
    model = applications.VGG16(include_top=False, weights='imagenet')                               # exclude 3 FC layers on top of network

    score_iou_btl_g, nb_btl_samples = get_images_count_recursive(dataset_btl_path)
    logging.debug('score_iou_btl_g {}'.format(score_iou_btl_g))
    logging.debug('nb_btl_samples {}'.format(nb_btl_samples))


    ## Train
    generator = datagen.flow_from_directory(
        dataset_btl_path,
        target_size=(img_width, img_height),
        batch_size=batch_size,
        classes=None,                                                                               #  the order of the classes, which will map to the label indices, will be alphanumeric
        class_mode=None,                                                                            # "categorical": 2D one-hot encoded labels; "None": yield batches of data, no labels; "sparse" will be 1D integer labels.
        save_to_dir='temp',
        shuffle=False)                                                                              # Don't shuffle else [class index = alphabetical folder order] logic used below might become wrong; first 1000 images will be cats, then 1000 dogs
    logging.info('generator.class_indices {}'.format(generator.class_indices))
                                                                                                    # classes: If not given, the order of the classes, which will map to the label indices, will be alphanumeric
    bottleneck_features_btl = model.predict_generator(
        generator, nb_btl_samples // batch_size)
    logging.debug('bottleneck_features_btl {}'.format(bottleneck_features_btl.shape))           # bottleneck_features_train (10534, 4, 4, 512) where train images i.e Blazer+Jeans=5408+5126=10532 images;

    # save the output as a Numpy array
    logging.debug('Saving bottleneck_features_btl...')
    np.save(open('output/bottleneck_features_btl.npy', 'w'),
            bottleneck_features_btl) 
Example #23
Source File: mlearn_for_image.py    From 12306 with MIT License 5 votes vote down vote up
def learn():
    (train_x, train_y, sample_weight), (test_x, test_y) = load_data()
    datagen = ImageDataGenerator(horizontal_flip=True,
                                 vertical_flip=True)
    train_generator = datagen.flow(train_x, train_y, sample_weight=sample_weight)
    base = VGG16(weights='imagenet', include_top=False, input_shape=(None, None, 3))
    for layer in base.layers[:-4]:
        layer.trainable = False
    model = models.Sequential([
        base,
        layers.BatchNormalization(),
        layers.Conv2D(64, (3, 3), activation='relu', padding='same'),
        layers.GlobalAveragePooling2D(),
        layers.BatchNormalization(),
        layers.Dense(64, activation='relu'),
        layers.BatchNormalization(),
        layers.Dropout(0.20),
        layers.Dense(80, activation='softmax')
    ])
    model.compile(optimizer=optimizers.RMSprop(lr=1e-5),
                  loss='sparse_categorical_crossentropy',
                  metrics=['accuracy'])
    model.summary()
    reduce_lr = ReduceLROnPlateau(verbose=1)
    model.fit_generator(train_generator, epochs=400,
                        steps_per_epoch=100,
                        validation_data=(test_x[:800], test_y[:800]),
                        callbacks=[reduce_lr])
    result = model.evaluate(test_x, test_y)
    print(result)
    model.save('12306.image.model.h5', include_optimizer=False) 
Example #24
Source File: features.py    From vergeml with MIT License 4 votes vote down vote up
def get_imagenet_architecture(architecture, variant, size, alpha, output_layer, include_top=False, weights='imagenet'):
    from keras import applications, Model

    if include_top:
        assert output_layer == 'last'

    if size == 'auto':
        size = get_image_size(architecture, variant, size)

    shape = (size, size, 3)

    if architecture == 'densenet':
        if variant == 'auto':
            variant = 'densenet-121'
        if variant == 'densenet-121':
            model = applications.DenseNet121(weights=weights, include_top=include_top, input_shape=shape)
        elif variant == 'densenet-169':
            model = applications.DenseNet169(weights=weights, include_top=include_top, input_shape=shape)
        elif variant == 'densenet-201':
            model = applications.DenseNet201(weights=weights, include_top=include_top, input_shape=shape)
    elif architecture == 'inception-resnet-v2':
        model = applications.InceptionResNetV2(weights=weights, include_top=include_top, input_shape=shape)
    elif architecture == 'mobilenet':
        model = applications.MobileNet(weights=weights, include_top=include_top, input_shape=shape, alpha=alpha)
    elif architecture == 'mobilenet-v2':
        model = applications.MobileNetV2(weights=weights, include_top=include_top, input_shape=shape, alpha=alpha)
    elif architecture == 'nasnet':
        if variant == 'auto':
            variant = 'large'
        if variant == 'large':
            model = applications.NASNetLarge(weights=weights, include_top=include_top, input_shape=shape)
        else:
            model = applications.NASNetMobile(weights=weights, include_top=include_top, input_shape=shape)
    elif architecture == 'resnet-50':
        model = applications.ResNet50(weights=weights, include_top=include_top, input_shape=shape)
    elif architecture == 'vgg-16':
        model = applications.VGG16(weights=weights, include_top=include_top, input_shape=shape)
    elif architecture == 'vgg-19':
        model = applications.VGG19(weights=weights, include_top=include_top, input_shape=shape)
    elif architecture == 'xception':
        model = applications.Xception(weights=weights, include_top=include_top, input_shape=shape)
    elif architecture == 'inception-v3':
        model = applications.InceptionV3(weights=weights, include_top=include_top, input_shape=shape)

    if output_layer != 'last':
        try:
            if isinstance(output_layer, int):
                layer = model.layers[output_layer]
            else:
                layer = model.get_layer(output_layer)
        except Exception:
            raise VergeMLError('layer not found: {}'.format(output_layer))
        model = Model(inputs=model.input, outputs=layer.output)

    return model 
Example #25
Source File: train_multi_v3.py    From DeepFashion with Apache License 2.0 4 votes vote down vote up
def create_model_predict(input_shape, optimizer='Adagrad', learn_rate=None, decay=0.0, momentum=0.0, activation='relu', dropout_rate=0.5):
    logging.debug('input_shape {}'.format(input_shape))
    logging.debug('input_shape {}'.format(type(input_shape)))

    # Optimizer
    optimizer, learn_rate = get_optimizer(optimizer, learn_rate, decay, momentum)

    input_shape = (img_width, img_height, 3)                                                        # 224,224,3
    base_model = applications.VGG16(weights='imagenet', include_top=False, input_shape=input_shape)
    logging.debug('base_model inputs {}'.format(base_model.input))                                  # shape=(?, 224, 224, 3)
    logging.debug('base_model outputs {}'.format(base_model.output))                                # shape=(?, 224, 224, 3)


    # TODO: Hardcoding
    input_shape_top_model_tensor = Input(shape=(7, 7, 512))
    #x_common = Dense(256, activation='relu')(input_shape_top_model)
    # x_common = Dense(256, activation='relu')(base_model.output)

    ## Model Classification
    # x = Flatten()(x_common)
    x = Flatten()(base_model.output)
    x = Dense(256, activation='tanh')(x)
    x = Dropout(dropout_rate)(x)
    predictions_class = Dense(len(class_names), activation='softmax', name='predictions_class')(x)


    ## Model (Regression) IOU score
    # x = Flatten()(x_common)
    x = Flatten()(base_model.output)
    x = Dense(256, activation='tanh')(x)
    x = Dropout(dropout_rate)(x)
    x = Dense(256, activation='tanh')(x)
    x = Dropout(dropout_rate)(x)
    predictions_iou = Dense(1, activation='sigmoid', name='predictions_iou')(x)

    # model_top.load_weights(top_model_weights_path)

    # This creates a model that includes the Input layer and three Dense layers
    logging.debug('Creating model {}')
    #model = Model(inputs=base_model.input, outputs=[predictions_class(base_model.output), predictions_iou(base_model.output)])
    model = Model(inputs=base_model.input, outputs=[predictions_class, predictions_iou])

    logging.debug('model summary {}'.format(model.summary()))

    # TODO: loads only top model weights as only those are present(also saved by name) in the file.
    # Test this assumption
    model.load_weights(top_model_weights_path, by_name=True)


    # Compile
    model.compile(optimizer=optimizer,
                  loss={'predictions_class': 'sparse_categorical_crossentropy', 'predictions_iou': 'mean_squared_error'}, metrics=['accuracy'],
                  loss_weights={'predictions_class': 0.5, 'predictions_iou': 0.5})
                  #loss={'predictions_class': 'sparse_categorical_crossentropy', 'predictions_iou': 'logcosh'}, metrics=['accuracy'],

    logging.info('optimizer:{}  learn_rate:{}  decay:{}  momentum:{}  activation:{}  dropout_rate:{}'.format(
        optimizer, learn_rate, decay, momentum, activation, dropout_rate))

    return model 
Example #26
Source File: train_multi_v3.py    From DeepFashion with Apache License 2.0 4 votes vote down vote up
def create_model_train(input_shape, optimizer='Adagrad', learn_rate=None, decay=0.0, momentum=0.0, activation='relu', dropout_rate=0.5):
    logging.debug('input_shape {}'.format(input_shape))
    logging.debug('input_shape {}'.format(type(input_shape)))

    # Optimizer
    optimizer, learn_rate = get_optimizer(optimizer, learn_rate, decay, momentum)

    # input_shape = (7, 7, 512)                                                                     # VGG bottleneck layer - block5_pool (MaxPooling2D)

    inputs = Input(shape=(input_shape))
    x_common = Dense(256, activation='relu')(inputs)

    ## Model Classification
    x = Flatten()(x_common)
    #x = Dropout(dropout_rate)(x)
    predictions_class = Dense(len(class_names), activation='softmax', name='predictions_class')(x)


    ## Model (Regression) IOU score
    x = Flatten()(x_common)
    # x = Dense(256, activation='relu')(x)
    # x = Dropout(dropout_rate)(x)
    predictions_iou = Dense(1, activation='sigmoid', name='predictions_iou')(x)


    # This creates a model that includes the Input layer and three Dense layers
    model = Model(inputs=inputs, outputs=[predictions_class, predictions_iou])


    logging.debug('model summary {}'.format(model.summary()))


    # Compile
    model.compile(optimizer=optimizer,
                  loss={'predictions_class': 'sparse_categorical_crossentropy', 'predictions_iou': 'mean_squared_error'}, metrics=['accuracy'],
                  loss_weights={'predictions_class': 0.5, 'predictions_iou': 0.5})
                  #loss={'predictions_class': 'sparse_categorical_crossentropy', 'predictions_iou': 'logcosh'}, metrics=['accuracy'],

    logging.info('optimizer:{}  learn_rate:{}  decay:{}  momentum:{}  activation:{}  dropout_rate:{}'.format(
        optimizer, learn_rate, decay, momentum, activation, dropout_rate))

    return model


# INPUT:
#           Input Image (None, 224, 224, 3) [fed to VGG16]
# OUTPUT:
#           Branch1 - Class Prediction
#           Branch2 - IOU Prediction

# NOTE: Both models in create_model_train() and  create_model_predict() should be exaclty same 
Example #27
Source File: train_multi_v3.py    From DeepFashion with Apache License 2.0 4 votes vote down vote up
def get_optimizer(optimizer='Adagrad', lr=None, decay=0.0, momentum=0.0):

    if optimizer == 'SGD':
        if lr is None:
            lr = 0.01
        optimizer_mod = keras.optimizers.SGD(lr=lr, momentum=momentum, decay=decay, nesterov=False)

    elif optimizer == 'RMSprop':
        if lr is None:
            lr = 0.001
        optimizer_mod = keras.optimizers.RMSprop(lr=lr, rho=0.9, epsilon=1e-08, decay=decay)

    elif optimizer == 'Adagrad':
        if lr is None:
            lr = 0.01
        optimizer_mod = keras.optimizers.Adagrad(lr=lr, epsilon=1e-08, decay=decay)

    elif optimizer == 'Adadelta':
        if lr is None:
            lr = 1.0
        optimizer_mod = keras.optimizers.Adadelta(lr=1.0, rho=0.95, epsilon=1e-08, decay=0.0)

    elif optimizer == 'Adam':
        if lr is None:
            lr = 0.001
        optimizer_mod = keras.optimizers.Adam(lr=0.001, beta_1=0.9, beta_2=0.999, epsilon=1e-08, decay=0.0)

    elif optimizer == 'Adamax':
        if lr is None:
            lr = 0.002
        optimizer_mod = keras.optimizers.Adamax(lr=0.002, beta_1=0.9, beta_2=0.999, epsilon=1e-08, decay=0.0)

    elif optimizer == 'Nadam':
        if lr is None:
            lr = 0.002
        optimizer_mod = keras.optimizers.Nadam(lr=0.002, beta_1=0.9, beta_2=0.999, epsilon=1e-08, schedule_decay=0.004)

    else:
        logging.error('Unknown optimizer {}'.format(optimizer))
        exit(1)

    # logging.debug('lr {}'.format(lr))
    # logging.debug('momentum {}'.format(momentum))
    # logging.debug('decay {}'.format(decay))
    # logging.debug('optimizer_mod {}'.format(optimizer_mod))

    return optimizer_mod, lr


# INPUT:
#           VGG16 - block5_pool (MaxPooling2D) (None, 7, 7, 512)
# OUTPUT:
#           Branch1 - Class Prediction
#           Branch2 - IOU Prediction

# NOTE: Both models in create_model_train() and  create_model_predict() should be exaclty same 
Example #28
Source File: train_multi_v4.py    From DeepFashion with Apache License 2.0 4 votes vote down vote up
def get_optimizer(optimizer='Adagrad', lr=None, decay=0.0, momentum=0.0):

    if optimizer == 'SGD':
        if lr is None:
            lr = 0.01
        optimizer_mod = keras.optimizers.SGD(lr=lr, momentum=momentum, decay=decay, nesterov=False)

    elif optimizer == 'RMSprop':
        if lr is None:
            lr = 0.001
        optimizer_mod = keras.optimizers.RMSprop(lr=lr, rho=0.9, epsilon=1e-08, decay=decay)

    elif optimizer == 'Adagrad':
        if lr is None:
            lr = 0.01
        optimizer_mod = keras.optimizers.Adagrad(lr=lr, epsilon=1e-08, decay=decay)

    elif optimizer == 'Adadelta':
        if lr is None:
            lr = 1.0
        optimizer_mod = keras.optimizers.Adadelta(lr=1.0, rho=0.95, epsilon=1e-08, decay=0.0)

    elif optimizer == 'Adam':
        if lr is None:
            lr = 0.001
        optimizer_mod = keras.optimizers.Adam(lr=0.001, beta_1=0.9, beta_2=0.999, epsilon=1e-08, decay=0.0)

    elif optimizer == 'Adamax':
        if lr is None:
            lr = 0.002
        optimizer_mod = keras.optimizers.Adamax(lr=0.002, beta_1=0.9, beta_2=0.999, epsilon=1e-08, decay=0.0)

    elif optimizer == 'Nadam':
        if lr is None:
            lr = 0.002
        optimizer_mod = keras.optimizers.Nadam(lr=0.002, beta_1=0.9, beta_2=0.999, epsilon=1e-08, schedule_decay=0.004)

    else:
        logging.error('Unknown optimizer {}'.format(optimizer))
        exit(1)

    # logging.debug('lr {}'.format(lr))
    # logging.debug('momentum {}'.format(momentum))
    # logging.debug('decay {}'.format(decay))
    # logging.debug('optimizer_mod {}'.format(optimizer_mod))

    return optimizer_mod, lr


# INPUT:
#           VGG16 - block5_pool (MaxPooling2D) (None, 7, 7, 512)
# OUTPUT:
#           Branch1 - Class Prediction
#           Branch2 - IOU Prediction

# NOTE: Both models in create_model_train() and  create_model_predict() should be exaclty same