Module tfhelper.transfer_learning.transfer_learning

Functions

def get_transfer_learning_model(target_model=<function VGG19>, input_shape=(32, 32, 3), weights='imagenet', n_class=10, optimizer='adam', loss='SparseCategoricalCrossentropy', metrics='accuracy', base_model_only=False)

Get TensorFlow models for transfer learning. Under progress. We do not recommend to use this function.

Args

target_model : tf.keras.models.Model

tf.keras.applications. …

input_shape : tuple

Input shape

weights : None, str

'imagenet'

n_class : int

Number of classes

optimizer : str, tf.keras.optimizers.Optimizer

Optimizer

loss : str, tf.keras.losses.Losses

Loss function

metrics : list of str

Metrics

base_model_only : bool

True - Not appending custom layer False - Appending custom layer

Returns:

Expand source code

def get_transfer_learning_model(target_model=tf.keras.applications.VGG19,
                                input_shape=(32, 32, 3), weights='imagenet', n_class=10,
                                optimizer='adam', loss='SparseCategoricalCrossentropy', metrics=('accuracy'),
                                base_model_only=False):
    """
    Get TensorFlow models for transfer learning.
    Under progress. We do not recommend to use this function.

    Args:
        target_model (tf.keras.models.Model): tf.keras.applications. ...
        input_shape (tuple): Input shape
        weights (None, str): 'imagenet'
        n_class (int): Number of classes
        optimizer (str, tf.keras.optimizers.Optimizer): Optimizer
        loss (str, tf.keras.losses.Losses): Loss function
        metrics (list of str): Metrics
        base_model_only (bool): True - Not appending custom layer
                                False - Appending custom layer

    Returns:

    """

    base_model_ = target_model(input_shape=input_shape, include_top=False, weights=weights)

    if weights is not None:
        base_model_.trainable = False

    if base_model_only:
        return base_model_

    try:
        model_ = tf.keras.Sequential(base_model_.layers + [
            tf.keras.layers.Flatten(),
            tf.keras.layers.Dense(n_class, activation='softmax')
        ])
    except ValueError:
        model_ = tf.keras.Sequential([base_model_,
                                     tf.keras.layers.Flatten(),
                                     tf.keras.layers.Dense(n_class, activation='softmax')
        ])

    model_.compile(optimizer=optimizer, loss=loss, metrics=metrics)

    return model_, base_model_