mediapipe_model_maker.image_classifier.HParams

The hyperparameters for training image classifiers.

Inherits From: BaseHParams

mediapipe_model_maker.image_classifier.HParams(
    learning_rate: float = 0.001,
    batch_size: int = 2,
    epochs: int = 10,
    steps_per_epoch: Optional[int] = None,
    class_weights: Optional[Mapping[int, float]] = None,
    shuffle: bool = False,
    repeat: bool = False,
    export_dir: str = tempfile.mkdtemp(),
    distribution_strategy: str = 'off',
    num_gpus: int = 0,
    tpu: str = '',
    do_fine_tuning: bool = False,
    l1_regularizer: float = 0.0,
    l2_regularizer: float = 0.0001,
    label_smoothing: float = 0.1,
    do_data_augmentation: bool = True,
    decay_samples: int = (10000 * 256),
    warmup_epochs: int = 2,
    checkpoint_frequency: int = 1,
    one_hot: bool = True,
    multi_labels: bool = False
)

learning_rate Learning rate to use for gradient descent training.
batch_size Batch size for training.
epochs Number of training iterations over the dataset.
do_fine_tuning If true, the base module is trained together with the classification layer on top.
l1_regularizer A regularizer that applies a L1 regularization penalty.
l2_regularizer A regularizer that applies a L2 regularization penalty.
label_smoothing Amount of label smoothing to apply. See tf.keras.losses for more details.
do_data_augmentation A boolean controlling whether the training dataset is augmented by randomly distorting input images, including random cropping, flipping, etc. See utils.image_preprocessing documentation for details.
decay_samples Number of training samples used to calculate the decay steps and create the training optimizer.
warmup_steps Number of warmup steps for a linear increasing warmup schedule on learning rate. Used to set up warmup schedule by model_util.WarmUp.
checkpoint_frequency Frequency to save checkpoint.
one_hot Whether the label data is score input or one-hot.
multi_labels Whether the model predict multi labels.
steps_per_epoch Dataclass field
class_weights Dataclass field
shuffle Dataclass field
repeat Dataclass field
export_dir Dataclass field
distribution_strategy Dataclass field
num_gpus Dataclass field
tpu Dataclass field
warmup_epochs Dataclass field

Methods

get_strategy

View source

get_strategy()

__eq__

__eq__(
    other
)

batch_size 2
checkpoint_frequency 1
class_weights None
decay_samples 2560000
distribution_strategy 'off'
do_data_augmentation True
do_fine_tuning False
epochs 10
export_dir '/tmpfs/tmp/tmpnt_h4p9w'
l1_regularizer 0.0
l2_regularizer 0.0001
label_smoothing 0.1
learning_rate 0.001
multi_labels False
num_gpus 0
one_hot True
repeat False
shuffle False
steps_per_epoch None
tpu ''
warmup_epochs 2