mediapipe_model_maker.object_detector.QATHParams

The hyperparameters for running quantization aware training (QAT) on object detectors.

mediapipe_model_maker.object_detector.QATHParams(
    learning_rate: float = 0.3,
    batch_size: int = 8,
    epochs: int = 15,
    decay_steps: int = 8,
    decay_rate: float = 0.96
)

For more information on QAT, see: https://www.tensorflow.org/model_optimization/guide/quantization/training

learning_rate Learning rate to use for gradient descent QAT.
batch_size Batch size for QAT.
epochs Number of training iterations over the dataset.
decay_steps Learning rate decay steps for Exponential Decay. See https://www.tensorflow.org/api_docs/python/tf/keras/optimizers/schedules/ExponentialDecay for more information.
decay_rate Learning rate decay rate for Exponential Decay. See https://www.tensorflow.org/api_docs/python/tf/keras/optimizers/schedules/ExponentialDecay for more information.

Methods

__eq__

__eq__(
    other
)

batch_size 8
decay_rate 0.96
decay_steps 8
epochs 15
learning_rate 0.3