set default as imagenet

README.md

@@ -36,7 +36,7 @@
     ```py
     pip install -U git+https://github.com/leondgarse/keras_efficientnet_v2
     ```
-  - **Define model and load pretrained weights** Parameter `pretrained` is added in value `[None, "imagenet", "imagenet21k", "imagenet21k-ft1k"]`, default is `imagenet21k-ft1k`.
+  - **Define model and load pretrained weights** Parameter `pretrained` is added in value `[None, "imagenet", "imagenet21k", "imagenet21k-ft1k"]`, default is `imagenet`.
     ```py
     # Will download and load `imagenet` pretrained weights.
     # Model weight is loaded with `by_name=True, skip_mismatch=True`.
@@ -142,24 +142,13 @@
   from tensorflow import keras
   from keras_efficientnet_v2 import progressive_train_test
 
-  num_classes = 10
-  ev2_s = keras_efficientnet_v2.EfficientNetV2("s", input_shape=(None, None, 3), num_classes=0)
-  out = ev2_s.output
-
-  nn = keras.layers.GlobalAveragePooling2D(name="avg_pool")(out)
-  nn = keras.layers.Dropout(0.1)(nn)
-  nn = keras.layers.Dense(num_classes, activation="softmax", name="predictions", dtype="float32")(nn)
-  model = keras.models.Model(ev2_s.inputs[0], nn)
-
-  lr_scheduler = None
-  optimizer = "adam"
-  loss = "categorical_crossentropy"
-  model.compile(loss=loss, optimizer=optimizer, metrics=["accuracy"])
+  model = keras_efficientnet_v2.EfficientNetV2S(input_shape=(None, None, 3), num_classes=10, classifier_activation='softmax', dropout=0.1)
+  model.compile(loss="categorical_crossentropy", optimizer="adam", metrics=["accuracy"])
 
   hhs = progressive_train_test.progressive_with_dropout_randaug(
       model,
       data_name="cifar10",
-      lr_scheduler=lr_scheduler,
+      lr_scheduler=None,
       total_epochs=36,
       batch_size=64,
       dropout_layer=-2,

keras_efficientnet_v2/efficientnet_v2.py

@@ -212,7 +212,7 @@ def EfficientNetV2(
     first_strides=2,
     drop_connect_rate=0,
     classifier_activation="softmax",
-    pretrained="imagenet21k-ft1k",
+    pretrained="imagenet",
     model_name="EfficientNetV2",
     kwargs=None,    # Not used, just recieving parameter
 ):
@@ -253,7 +253,7 @@ def EfficientNetV2(
         nn = GlobalAveragePooling2D(name="avg_pool")(nn)
         if dropout > 0 and dropout < 1:
             nn = Dropout(dropout)(nn)
-        nn = Dense(num_classes, activation=classifier_activation, name="predictions")(nn)
+        nn = Dense(num_classes, activation=classifier_activation, dtype="float32", name="predictions")(nn)
 
     model = Model(inputs=inputs, outputs=nn, name=model_name)
     reload_model_weights(model, model_type, pretrained)
@@ -279,39 +279,39 @@ def reload_model_weights(model, model_type, pretrained="imagenet"):
         model.load_weights(pretrained_model, by_name=True, skip_mismatch=True)
 
 
-def EfficientNetV2B0(input_shape=(224, 224, 3), num_classes=1000, dropout=0.2, classifier_activation="softmax", pretrained="imagenet21k-ft1k", **kwargs):
+def EfficientNetV2B0(input_shape=(224, 224, 3), num_classes=1000, dropout=0.2, classifier_activation="softmax", pretrained="imagenet", **kwargs):
     return EfficientNetV2(model_type="b0", model_name="EfficientNetV2B0", **locals(), **kwargs)
 
 
-def EfficientNetV2B1(input_shape=(240, 240, 3), num_classes=1000, dropout=0.2, classifier_activation="softmax", pretrained="imagenet21k-ft1k", **kwargs):
+def EfficientNetV2B1(input_shape=(240, 240, 3), num_classes=1000, dropout=0.2, classifier_activation="softmax", pretrained="imagenet", **kwargs):
     return EfficientNetV2(model_type="b1", model_name="EfficientNetV2B1", **locals(), **kwargs)
 
 
-def EfficientNetV2B2(input_shape=(260, 260, 3), num_classes=1000, dropout=0.3, classifier_activation="softmax", pretrained="imagenet21k-ft1k", **kwargs):
+def EfficientNetV2B2(input_shape=(260, 260, 3), num_classes=1000, dropout=0.3, classifier_activation="softmax", pretrained="imagenet", **kwargs):
     return EfficientNetV2(model_type="b2", model_name="EfficientNetV2B2", **locals(), **kwargs)
 
 
-def EfficientNetV2B3(input_shape=(300, 300, 3), num_classes=1000, dropout=0.3, classifier_activation="softmax", pretrained="imagenet21k-ft1k", **kwargs):
+def EfficientNetV2B3(input_shape=(300, 300, 3), num_classes=1000, dropout=0.3, classifier_activation="softmax", pretrained="imagenet", **kwargs):
     return EfficientNetV2(model_type="b3", model_name="EfficientNetV2B3", **locals(), **kwargs)
 
 
 def EfficientNetV2T(input_shape=(320, 320, 3), num_classes=1000, dropout=0.2, classifier_activation="softmax", pretrained="imagenet", **kwargs):
     return EfficientNetV2(model_type="t", model_name="EfficientNetV2T", **locals(), **kwargs)
 
 
-def EfficientNetV2S(input_shape=(384, 384, 3), num_classes=1000, dropout=0.2, classifier_activation="softmax", pretrained="imagenet21k-ft1k", **kwargs):
+def EfficientNetV2S(input_shape=(384, 384, 3), num_classes=1000, dropout=0.2, classifier_activation="softmax", pretrained="imagenet", **kwargs):
     return EfficientNetV2(model_type="s", model_name="EfficientNetV2S", **locals(), **kwargs)
 
 
-def EfficientNetV2M(input_shape=(480, 480, 3), num_classes=1000, dropout=0.3, classifier_activation="softmax", pretrained="imagenet21k-ft1k", **kwargs):
+def EfficientNetV2M(input_shape=(480, 480, 3), num_classes=1000, dropout=0.3, classifier_activation="softmax", pretrained="imagenet", **kwargs):
     return EfficientNetV2(model_type="m", model_name="EfficientNetV2M", **locals(), **kwargs)
 
 
-def EfficientNetV2L(input_shape=(480, 480, 3), num_classes=1000, dropout=0.4, classifier_activation="softmax", pretrained="imagenet21k-ft1k", **kwargs):
+def EfficientNetV2L(input_shape=(480, 480, 3), num_classes=1000, dropout=0.4, classifier_activation="softmax", pretrained="imagenet", **kwargs):
     return EfficientNetV2(model_type="l", model_name="EfficientNetV2L", **locals(), **kwargs)
 
 
-def EfficientNetV2XL(input_shape=(512, 512, 3), num_classes=1000, dropout=0.4, classifier_activation="softmax", pretrained="imagenet21k-ft1k", **kwargs):
+def EfficientNetV2XL(input_shape=(512, 512, 3), num_classes=1000, dropout=0.4, classifier_activation="softmax", pretrained="imagenet", **kwargs):
     return EfficientNetV2(model_type="xl", model_name="EfficientNetV2XL", **locals(), **kwargs)