segmodels-keras 0.1.0.dev2__tar.gz → 0.1.0.dev4__tar.gz

{segmodels_keras-0.1.0.dev2 → segmodels_keras-0.1.0.dev4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: segmodels_keras
-Version: 0.1.0.dev2
+Version: 0.1.0.dev4
 Summary: Image segmentation models with pre-trained backbones with Keras.
 Home-page: https://github.com/orthoseg/segmodels_keras
 Author: Pieter Roggemans

{segmodels_keras-0.1.0.dev2 → segmodels_keras-0.1.0.dev4}/segmodels_keras/__init__.py

@@ -2,9 +2,9 @@ import functools  # noqa: I001
 import os
 
 # __version__ should be defined as soon as possible
-from segmodels_keras._version import __version__
+from ._version import __version__
 
-from segmodels_keras import base
+from . import base
 
 _KERAS_FRAMEWORK_NAME = "keras"
 _TF_KERAS_FRAMEWORK_NAME = "tf.keras"

{segmodels_keras-0.1.0.dev2 → segmodels_keras-0.1.0.dev4}/segmodels_keras/_version.py

@@ -1,2 +1,2 @@
-__versioninfo__ = (0, 1, 0, "dev2")
+__versioninfo__ = (0, 1, 0, "dev4")
 __version__ = ".".join(map(str, __versioninfo__))

{segmodels_keras-0.1.0.dev2 → segmodels_keras-0.1.0.dev4}/segmodels_keras/backbones/backbones_factory.py

@@ -125,14 +125,16 @@ class BackbonesFactory:
         "inceptionresnetv2": (
             irv2.InceptionResNetV2,
             irv2.preprocess_input,
-            (
-                "activation_161",
-                "activation_74",
-                "activation_3",
-                "activation",
-                # "input_1",
-            ),
-            # (594, 260, 16, 9),
+            # Use the layer indexes instead of names because otherwise loading weights
+            # saved with keras 2 cannot be loaded with keras 3.
+            # (
+            #     "activation_161",
+            #     "activation_74",
+            #     "activation_3",
+            #     "activation",
+            #     # "input_1",
+            # ),
+            (594, 260, 16, 9),
         ),
         "inceptionv3": (iv3.InceptionV3, iv3.preprocess_input, (228, 86, 16, 9)),
         # MobileNet
@@ -272,5 +274,10 @@ class BackbonesFactory:
     def get_preprocessing(self, name):
         return self._models.get(name)[1]
 
+    def get_custom_objects(self, name):
+        if name == "inceptionresnetv2":
+            return irv2.get_custom_objects()
+        return {}
+
 
 Backbones = BackbonesFactory()

{segmodels_keras-0.1.0.dev2 → segmodels_keras-0.1.0.dev4}/segmodels_keras/backbones/inception_resnet_v2.py

@@ -13,7 +13,11 @@ https://github.com/tensorflow/models/tree/master/research/slim#pre-trained-model
 - [Inception-v4, Inception-ResNet and the Impact of
    Residual Connections on Learning](https://arxiv.org/abs/1602.07261) (AAAI 2017)
 
-Differences with the original implementation in keras.applications:
+The differences retained from the segmentation_models version
+(https://github.com/qubvel/segmentation_models/blob/master/segmentation_models/backbones/inception_resnet_v2.py)
+compared to the most recent version in keras.applications
+(https://github.com/keras-team/keras/blob/86d7deacb0ac2baa3e41db4bcd61775065b5a301/keras/src/applications/inception_resnet_v2.py)
+are the following:
 - For e.g. in the `conv2d_bn` blocks, padding="same" is used instead of "valid". This is
   to make the sizes of the outputs of the layers the correct size for the skip
   connections for the decoders in the segmentation models. It does not affect the
@@ -24,11 +28,8 @@ Differences with the original implementation in keras.applications:
   also give issues when loading weights that were saved with an older version. An
   example where this solution is used/shown can be found here:
   https://github.com/ayushdabra/dubai-satellite-imagery-segmentation/blob/9446e16134e752dda080ba9d65be18ea47fa26a9/get_activations.py#L40
-- The original implementation uses different imports as it is internal in keras, and
-  here we need to use the public API of keras.
-- Use Lambda layer instead of CustomScaleLayer because:
-    - it avoids having to define + pass in a custom layer when loading the model
-    - otherwise "old" weights cannot be loaded when using keras 3
+- The keras.applications implementation uses different imports as it is internal in
+  keras, and here we need to use the public API of keras.
 """
 
 import warnings
@@ -36,7 +37,6 @@ from pathlib import Path
 
 from keras import backend, layers, models
 from keras.applications import imagenet_utils
-from keras.layers import Layer
 from keras.utils import get_file, get_source_inputs
 
 BASE_WEIGHT_URL = (
@@ -311,6 +311,20 @@ def conv2d_bn(
     return x
 
 
+class CustomScaleLayer(layers.Layer):
+    def __init__(self, scale, **kwargs):
+        super().__init__(**kwargs)
+        self.scale = scale
+
+    def get_config(self):
+        config = super().get_config()
+        config.update({"scale": self.scale})
+        return config
+
+    def call(self, inputs):
+        return inputs[0] + inputs[1] * self.scale
+
+
 def inception_resnet_block(x, scale, block_type, block_idx, activation="relu"):
     """Adds a Inception-ResNet block.
 
@@ -387,12 +401,7 @@ def inception_resnet_block(x, scale, block_type, block_idx, activation="relu"):
         name=f"{block_name}_conv",
     )
 
-    x = layers.Lambda(
-        lambda inputs, scale: inputs[0] + inputs[1] * scale,
-        output_shape=x.shape[1:],
-        arguments={"scale": scale},
-        name=block_name,
-    )([x, up])
+    x = CustomScaleLayer(scale)([x, up])
 
     if activation is not None:
         x = layers.Activation(activation, name=f"{block_name}_ac")(x)
@@ -431,6 +440,10 @@ def decode_predictions(preds, top=5):
     return imagenet_utils.decode_predictions(preds, top=top)
 
 
+def get_custom_objects():
+    return {"CustomScaleLayer": CustomScaleLayer}
+
+
 def _obtain_input_shape(
     input_shape,
     default_size,

{segmodels_keras-0.1.0.dev2 → segmodels_keras-0.1.0.dev4}/segmodels_keras.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: segmodels_keras
-Version: 0.1.0.dev2
+Version: 0.1.0.dev4
 Summary: Image segmentation models with pre-trained backbones with Keras.
 Home-page: https://github.com/orthoseg/segmodels_keras
 Author: Pieter Roggemans