Augmentation layer choice between trainings.

Author: Arkkienkeli

plugins/models/efficientnet.py

@@ -4,6 +4,7 @@
 import efficientnet.tfkeras as efn
 
 from deepprofiler.imaging.augmentations import AugmentationLayer
+from deepprofiler.imaging.augmentations import AugmentationLayerV2
 from deepprofiler.learning.model import DeepProfilerModel
 from deepprofiler.learning.tf2train import DeepProfilerModelV2
 
@@ -12,12 +13,16 @@ def model_factory(config, dset, crop_generator, val_crop_generator, is_training)
     if inspect.currentframe().f_back.f_code.co_name == 'learn_model_v2':
         tf.compat.v1.enable_v2_behavior()
         tf.config.run_functions_eagerly(True)
-        return createModelClass(DeepProfilerModelV2, config, dset, crop_generator, val_crop_generator, is_training)
+        augmentation_base = AugmentationLayerV2()
+        return createModelClass(DeepProfilerModelV2, config, dset, crop_generator,
+                                val_crop_generator, is_training, augmentation_base)
     else:
-        return createModelClass(DeepProfilerModel, config, dset, crop_generator, val_crop_generator, is_training)
+        augmentation_base = AugmentationLayer()
+        return createModelClass(DeepProfilerModel, config, dset, crop_generator,
+                                val_crop_generator, is_training, augmentation_base)
 
 
-def createModelClass(base, config, dset, crop_generator, val_crop_generator, is_training):
+def createModelClass(base, config, dset, crop_generator, val_crop_generator, is_training, augmentation_base):
     class ModelClass(base):
         def __init__(self, config, dset, crop_generator, val_crop_generator, is_training):
             super(ModelClass, self).__init__(config, dset, crop_generator, val_crop_generator, is_training)
@@ -99,7 +104,7 @@ def define_model(self, config, dset):
                 if self.config["train"]["model"].get("augmentations") is True:
                     model = tf.compat.v1.keras.models.model_from_json(
                         model.to_json(),
-                        {'AugmentationLayer': AugmentationLayer}
+                        {'AugmentationLayer': augmentation_base}
                     )
                 else:
                     model = tf.compat.v1.keras.models.model_from_json(model.to_json())
@@ -108,7 +113,7 @@ def define_model(self, config, dset):
 
         def copy_pretrained_weights(self):
             base_model = self.get_model(self.config, weights="imagenet")
-            lshift = self.feature_model.layers[1].name == 'augmentation_layer'  # Shift one layer to accommodate the AugmentationLayer
+            lshift = self.feature_model.layers[1].name == 'augmentation_layer_1'  # Shift one layer to accommodate the AugmentationLayer
 
             # => Transfer all weights except conv1.1
             total_layers = len(base_model.layers)

plugins/models/resnet.py

@@ -5,6 +5,7 @@
 from deepprofiler.learning.model import DeepProfilerModel
 from deepprofiler.learning.tf2train import DeepProfilerModelV2
 from deepprofiler.imaging.augmentations import AugmentationLayer
+from deepprofiler.imaging.augmentations import AugmentationLayerV2
 
 
 ##################################################
@@ -19,12 +20,16 @@ def model_factory(config, dset, crop_generator, val_crop_generator, is_training)
     if inspect.currentframe().f_back.f_code.co_name == 'learn_model_v2':
         tf.compat.v1.enable_v2_behavior()
         tf.config.run_functions_eagerly(True)
-        return createModelClass(DeepProfilerModelV2, config, dset, crop_generator, val_crop_generator, is_training)
+        augmentation_base = AugmentationLayerV2()
+        return createModelClass(DeepProfilerModelV2, config, dset, crop_generator,
+                                val_crop_generator, is_training, augmentation_base)
     else:
-        return createModelClass(DeepProfilerModel, config, dset, crop_generator, val_crop_generator, is_training)
+        augmentation_base = AugmentationLayer()
+        return createModelClass(DeepProfilerModel, config, dset, crop_generator,
+                                val_crop_generator, is_training, augmentation_base)
 
 
-def createModelClass(base, config, dset, crop_generator, val_crop_generator, is_training):
+def createModelClass(base, config, dset, crop_generator, val_crop_generator, is_training, augmentation_base):
     class ModelClass(base):
         def __init__(self, config, dset, crop_generator, val_crop_generator, is_training):
             super(ModelClass, self).__init__(config, dset, crop_generator, val_crop_generator, is_training)
@@ -105,7 +110,7 @@ def define_model(self, config, dset):
                 if self.config["train"]["model"].get("augmentations") is True:
                     model = tf.compat.v1.keras.models.model_from_json(
                         model.to_json(),
-                        {'AugmentationLayer': AugmentationLayer}
+                        {'AugmentationLayer': augmentation_base}
                     )
                 else:
                     model = tf.compat.v1.keras.models.model_from_json(model.to_json())
@@ -117,7 +122,7 @@ def define_model(self, config, dset):
         ## Support for ImageNet initialization
         def copy_pretrained_weights(self):
             base_model = self.get_model(self.config, weights="imagenet")
-            lshift = self.feature_model.layers[1].name == 'augmentation_layer'  # Shift one layer to accommodate the AugmentationLayer
+            lshift = self.feature_model.layers[1].name == 'augmentation_layer_1'  # Shift one layer to accommodate the AugmentationLayer
 
             # => Transfer all weights except conv1.1
             total_layers = len(base_model.layers)