Model Maker object detector change learning_rate_boundaries to learning_rate_epoch_boundaries.

PiperOrigin-RevId: 521024056

mediapipe/model_maker/python/vision/object_detector/hyperparameters.py

@@ -29,9 +29,9 @@ class HParams(hp.BaseHParams):
     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.
-    learning_rate_boundaries: List of epoch boundaries where
+    learning_rate_epoch_boundaries: List of epoch boundaries where
-      learning_rate_boundaries[i] is the epoch where the learning rate will
+      learning_rate_epoch_boundaries[i] is the epoch where the learning rate
-      decay to learning_rate * learning_rate_decay_multipliers[i].
+      will decay to learning_rate * learning_rate_decay_multipliers[i].
     learning_rate_decay_multipliers: List of learning rate multipliers which
       calculates the learning rate at the ith boundary as learning_rate *
       learning_rate_decay_multipliers[i].
@@ -43,35 +43,39 @@ class HParams(hp.BaseHParams):
   epochs: int = 10
 
   # Parameters for learning rate decay
-  learning_rate_boundaries: List[int] = dataclasses.field(
+  learning_rate_epoch_boundaries: List[int] = dataclasses.field(
-      default_factory=lambda: [5, 8]
+      default_factory=lambda: []
   )
   learning_rate_decay_multipliers: List[float] = dataclasses.field(
-      default_factory=lambda: [0.1, 0.01]
+      default_factory=lambda: []
   )
 
   def __post_init__(self):
     # Validate stepwise learning rate parameters
-    lr_boundary_len = len(self.learning_rate_boundaries)
+    lr_boundary_len = len(self.learning_rate_epoch_boundaries)
     lr_decay_multipliers_len = len(self.learning_rate_decay_multipliers)
     if lr_boundary_len != lr_decay_multipliers_len:
       raise ValueError(
-          "Length of learning_rate_boundaries and ",
+          "Length of learning_rate_epoch_boundaries and ",
           "learning_rate_decay_multipliers do not match: ",
           f"{lr_boundary_len}!={lr_decay_multipliers_len}",
       )
-    # Validate learning_rate_boundaries
+    # Validate learning_rate_epoch_boundaries
-    if sorted(self.learning_rate_boundaries) != self.learning_rate_boundaries:
-      raise ValueError(
-          "learning_rate_boundaries is not in ascending order: ",
-          self.learning_rate_boundaries,
-      )
     if (
-        self.learning_rate_boundaries
+        sorted(self.learning_rate_epoch_boundaries)
-        and self.learning_rate_boundaries[-1] > self.epochs
+        != self.learning_rate_epoch_boundaries
     ):
       raise ValueError(
-          "Values in learning_rate_boundaries cannot be greater ", "than epochs"
+          "learning_rate_epoch_boundaries is not in ascending order: ",
+          self.learning_rate_epoch_boundaries,
+      )
+    if (
+        self.learning_rate_epoch_boundaries
+        and self.learning_rate_epoch_boundaries[-1] > self.epochs
+    ):
+      raise ValueError(
+          "Values in learning_rate_epoch_boundaries cannot be greater ",
+          "than epochs",
       )
 
 

mediapipe/model_maker/python/vision/object_detector/object_detector.py

@@ -57,7 +57,6 @@ class ObjectDetector(classifier.Classifier):
     self._preprocessor = preprocessor.Preprocessor(model_spec)
     self._hparams = hparams
     self._model_options = model_options
-    self._optimizer = self._create_optimizer()
     self._is_qat = False
 
   @classmethod
@@ -104,6 +103,11 @@ class ObjectDetector(classifier.Classifier):
       train_data: Training data.
       validation_data: Validation data.
     """
+    self._optimizer = self._create_optimizer(
+        model_util.get_steps_per_epoch(
+            self._hparams.steps_per_epoch,
+        )
+    )
     self._create_model()
     self._train_model(
         train_data, validation_data, preprocessor=self._preprocessor
@@ -333,21 +337,34 @@ class ObjectDetector(classifier.Classifier):
     with open(metadata_file, 'w') as f:
       f.write(metadata_json)
 
-  def _create_optimizer(self) -> tf.keras.optimizers.Optimizer:
+  def _create_optimizer(
+      self, steps_per_epoch: int
+  ) -> tf.keras.optimizers.Optimizer:
     """Creates an optimizer with learning rate schedule for regular training.
 
     Uses Keras PiecewiseConstantDecay schedule by default.
 
+    Args:
+      steps_per_epoch: Steps per epoch to calculate the step boundaries from the
+        learning_rate_epoch_boundaries
+
     Returns:
       A tf.keras.optimizer.Optimizer for model training.
     """
     init_lr = self._hparams.learning_rate * self._hparams.batch_size / 256
+    if self._hparams.learning_rate_epoch_boundaries:
       lr_values = [init_lr] + [
           init_lr * m for m in self._hparams.learning_rate_decay_multipliers
       ]
-    learning_rate_fn = tf.keras.optimizers.schedules.PiecewiseConstantDecay(
+      lr_step_boundaries = [
-        self._hparams.learning_rate_boundaries, lr_values
+          steps_per_epoch * epoch_boundary
+          for epoch_boundary in self._hparams.learning_rate_epoch_boundaries
+      ]
+      learning_rate = tf.keras.optimizers.schedules.PiecewiseConstantDecay(
+          lr_step_boundaries, lr_values
       )
+    else:
+      learning_rate = init_lr
     return tf.keras.optimizers.experimental.SGD(
-        learning_rate=learning_rate_fn, momentum=0.9
+        learning_rate=learning_rate, momentum=0.9
     )