segmodels-keras 0.1.0.dev11__tar.gz → 0.1.0.dev13__tar.gz

{segmodels_keras-0.1.0.dev11 → segmodels_keras-0.1.0.dev13}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: segmodels_keras
-Version: 0.1.0.dev11
+Version: 0.1.0.dev13
 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.dev11 → segmodels_keras-0.1.0.dev13}/pyproject.toml

@@ -51,6 +51,7 @@ ignore = [
     ### Intentionally disabled
     "E402",     # Module level import not at top of file
     "PLR0913",  # Too many arguments to function call
+    "PLR0917",  # Too many positional arguments
     "PLR0911",  # Too many returns
     "PLR0912",  # Too many branches
     "PLR0915",  # Too many statements

segmodels_keras-0.1.0.dev13/segmodels_keras/_version.py

@@ -0,0 +1,2 @@
+__versioninfo__: tuple[int, int, int, str] = (0, 1, 0, "dev13")
+__version__: str = ".".join(map(str, __versioninfo__))

segmodels_keras-0.1.0.dev13/segmodels_keras/backbones/backbones_factory.py

@@ -0,0 +1,423 @@
+"""Backbones factory module."""
+
+import keras
+import keras.applications as ka
+
+from . import inception_resnet_v2 as irv2
+from . import inception_v3 as iv3
+from . import resnet_18_34
+
+
+class BackbonesFactory:
+    """Dict with all supported backbones.
+
+    Each backbone is represented as a tuple of 3 elements:
+    1. Backbone class
+    2. Preprocessing function
+    3. List of layers to take features from backbone in the following order:
+       (x16, x8, x4, x2, x1) - `x4` mean that features has 4 times less spatial
+       resolution (Height x Width) than input image.
+    """
+
+    _models = {  # noqa: RUF012
+        # Feature sizes in comments are HxWxC for a 512x512 input.
+        # ResNets < 50 layers are NOT available via keras.applications, so use a
+        # torchvision-compatible post-activation BasicBlock implementation.
+        # Layer names follow the keras.applications ResNet50+ naming convention.
+        "resnet18": (
+            resnet_18_34.ResNet18,
+            resnet_18_34.preprocess_input,
+            # resnet18: layer1=2, layer2=2, layer3=2, layer4=2 blocks
+            # backbone output = conv5_block2_out (512ch); skips are the 4 stages before
+            (
+                "conv4_block2_out",  # 32x32x256
+                "conv3_block2_out",  # 64x64x128
+                "conv2_block2_out",  # 128x128x64
+                "conv1_relu",  # 256x256x64
+            ),
+        ),
+        "resnet34": (
+            resnet_18_34.ResNet34,
+            resnet_18_34.preprocess_input,
+            # resnet34: layer1=3, layer2=4, layer3=6, layer4=3 blocks
+            # backbone output = conv5_block3_out (512ch); skips are the 4 stages before
+            (
+                "conv4_block6_out",  # 32x32x256
+                "conv3_block4_out",  # 64x64x128
+                "conv2_block3_out",  # 128x128x64
+                "conv1_relu",  # 256x256x64
+            ),
+        ),
+        # ResNets > 50 layers are available via keras.applications, so use those.
+        # Skip layers (inverted) from https://github.com/yingkaisha/keras-unet-collection
+        "resnet50": (
+            ka.ResNet50,
+            ka.resnet.preprocess_input,
+            (
+                "conv4_block6_out",  # 32x32x1024
+                "conv3_block4_out",  # 64x64x512
+                "conv2_block3_out",  # 128x128x256
+                "conv1_relu",  # 256x256x64
+            ),
+        ),
+        "resnet101": (
+            ka.ResNet101,
+            ka.resnet.preprocess_input,
+            (
+                "conv4_block23_out",  # 32x32x1024
+                "conv3_block4_out",  # 64x64x512
+                "conv2_block3_out",  # 128x128x256
+                "conv1_relu",  # 256x256x64
+            ),
+        ),
+        "resnet152": (
+            ka.ResNet152,
+            ka.resnet.preprocess_input,
+            (
+                "conv4_block36_out",  # 32x32x1024
+                "conv3_block8_out",  # 64x64x512
+                "conv2_block3_out",  # 128x128x256
+                "conv1_relu",  # 256x256x64
+            ),
+        ),
+        # ResNetV2
+        # Skip layers (inverted) from https://github.com/yingkaisha/keras-unet-collection
+        "resnet50v2": (
+            ka.ResNet50V2,
+            ka.resnet_v2.preprocess_input,
+            (
+                "conv4_block6_1_relu",  # 32x32x256
+                "conv3_block4_1_relu",  # 64x64x128
+                "conv2_block3_1_relu",  # 128x128x64
+                "conv1_conv",  # 256x256x64
+            ),
+        ),
+        "resnet101v2": (
+            ka.ResNet101V2,
+            ka.resnet_v2.preprocess_input,
+            (
+                "conv4_block23_1_relu",  # 32x32x256
+                "conv3_block4_1_relu",  # 64x64x128
+                "conv2_block3_1_relu",  # 128x128x64
+                "conv1_conv",  # 256x256x64
+            ),
+        ),
+        "resnet152v2": (
+            ka.ResNet152V2,
+            ka.resnet_v2.preprocess_input,
+            (
+                "conv4_block36_1_relu",  # 32x32x256
+                "conv3_block8_1_relu",  # 64x64x128
+                "conv2_block3_1_relu",  # 128x128x64
+                "conv1_conv",  # 256x256x64
+            ),
+        ),
+        # VGG
+        # Skip layers from segmentation_models
+        "vgg16": (
+            ka.vgg16.VGG16,
+            ka.vgg16.preprocess_input,
+            (
+                "block5_conv3",  # 32x32x512
+                "block4_conv3",  # 64x64x512
+                "block3_conv3",  # 128x128x256
+                "block2_conv2",  # 256x256x128
+                "block1_conv2",  # 512x512x64
+            ),
+        ),
+        "vgg19": (
+            ka.vgg19.VGG19,
+            ka.vgg19.preprocess_input,
+            (
+                "block5_conv4",  # 32x32x512
+                "block4_conv4",  # 64x64x512
+                "block3_conv4",  # 128x128x256
+                "block2_conv2",  # 256x256x128
+                "block1_conv2",  # 512x512x64
+            ),
+        ),
+        # DenseNet
+        # Skip layers (inverted) from https://github.com/yingkaisha/keras-unet-collection
+        "densenet121": (
+            ka.densenet.DenseNet121,
+            ka.densenet.preprocess_input,
+            (
+                "pool4_conv",  # 32x32x512
+                "pool3_conv",  # 64x64x256
+                "pool2_conv",  # 128x128x128
+                "conv1/relu",  # 256x256x64
+            ),
+        ),
+        "densenet169": (
+            ka.densenet.DenseNet169,
+            ka.densenet.preprocess_input,
+            (
+                "pool4_conv",  # 32x32x640
+                "pool3_conv",  # 64x64x256
+                "pool2_conv",  # 128x128x128
+                "conv1/relu",  # 256x256x64
+            ),
+        ),
+        "densenet201": (
+            ka.densenet.DenseNet201,
+            ka.densenet.preprocess_input,
+            (
+                "pool4_conv",  # 32x32x896
+                "pool3_conv",  # 64x64x256
+                "pool2_conv",  # 128x128x128
+                "conv1/relu",  # 256x256x64
+            ),
+        ),
+        # Inception
+        # Skip layers from segmentation_models
+        "inceptionresnetv2": (
+            irv2.InceptionResNetV2,
+            irv2.preprocess_input,
+            # 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,  # 32x32x1088
+                260,  # 64x64x320
+                16,  # 128x128x192
+                9,  # 256x256x64
+            ),
+        ),
+        "inceptionv3": (
+            iv3.InceptionV3,
+            iv3.preprocess_input,
+            (
+                228,  # 32x32x768
+                86,  # 64x64x288
+                16,  # 128x128x192
+                9,  # 256x256x64
+            ),
+        ),
+        # MobileNet
+        # Skip layers from segmentation_models
+        "mobilenet": (
+            ka.mobilenet.MobileNet,
+            ka.mobilenet.preprocess_input,
+            (
+                "conv_pw_11_relu",  # 32x32x512
+                "conv_pw_5_relu",  # 64x64x256
+                "conv_pw_3_relu",  # 128x128x128
+                "conv_pw_1_relu",  # 256x256x64
+            ),
+        ),
+        "mobilenetv2": (
+            ka.mobilenet_v2.MobileNetV2,
+            ka.mobilenet_v2.preprocess_input,
+            (
+                "block_13_expand_relu",  # 32x32x576
+                "block_6_expand_relu",  # 64x64x192
+                "block_3_expand_relu",  # 128x128x144
+                "block_1_expand_relu",  # 256x256x96
+            ),
+        ),
+        # EfficientNet
+        # Skip layers from segmentation_models
+        "efficientnetb0": [
+            ka.EfficientNetB0,
+            ka.efficientnet.preprocess_input,
+            (
+                "block6a_expand_activation",  # 32x32x672
+                "block4a_expand_activation",  # 64x64x240
+                "block3a_expand_activation",  # 128x128x144
+                "block2a_expand_activation",  # 256x256x96
+            ),
+        ],
+        "efficientnetb1": [
+            ka.EfficientNetB1,
+            ka.efficientnet.preprocess_input,
+            (
+                "block6a_expand_activation",  # 32x32x672
+                "block4a_expand_activation",  # 64x64x240
+                "block3a_expand_activation",  # 128x128x144
+                "block2a_expand_activation",  # 256x256x96
+            ),
+        ],
+        "efficientnetb2": [
+            ka.EfficientNetB2,
+            ka.efficientnet.preprocess_input,
+            (
+                "block6a_expand_activation",  # 32x32x720
+                "block4a_expand_activation",  # 64x64x288
+                "block3a_expand_activation",  # 128x128x144
+                "block2a_expand_activation",  # 256x256x96
+            ),
+        ],
+        "efficientnetb3": [
+            ka.EfficientNetB3,
+            ka.efficientnet.preprocess_input,
+            (
+                "block6a_expand_activation",  # 32x32x816
+                "block4a_expand_activation",  # 64x64x288
+                "block3a_expand_activation",  # 128x128x192
+                "block2a_expand_activation",  # 256x256x144
+            ),
+        ],
+        "efficientnetb4": [
+            ka.EfficientNetB4,
+            ka.efficientnet.preprocess_input,
+            (
+                "block6a_expand_activation",  # 32x32x960
+                "block4a_expand_activation",  # 64x64x336
+                "block3a_expand_activation",  # 128x128x192
+                "block2a_expand_activation",  # 256x256x144
+            ),
+        ],
+        "efficientnetb5": [
+            ka.EfficientNetB5,
+            ka.efficientnet.preprocess_input,
+            (
+                "block6a_expand_activation",  # 32x32x1056
+                "block4a_expand_activation",  # 64x64x384
+                "block3a_expand_activation",  # 128x128x240
+                "block2a_expand_activation",  # 256x256x144
+            ),
+        ],
+        "efficientnetb6": [
+            ka.EfficientNetB6,
+            ka.efficientnet.preprocess_input,
+            (
+                "block6a_expand_activation",  # 32x32x1200
+                "block4a_expand_activation",  # 64x64x432
+                "block3a_expand_activation",  # 128x128x240
+                "block2a_expand_activation",  # 256x256x192
+            ),
+        ],
+        "efficientnetb7": [
+            ka.EfficientNetB7,
+            ka.efficientnet.preprocess_input,
+            (
+                "block6a_expand_activation",  # 32x32x1344
+                "block4a_expand_activation",  # 64x64x480
+                "block3a_expand_activation",  # 128x128x288
+                "block2a_expand_activation",  # 256x256x192
+            ),
+        ],
+        # EfficientNetV2
+        # Stage layout differs from EfficientNet-Bx.
+        # Skip layers selected to match decoder resolutions:
+        # x16 -> x8 -> x4 -> x2
+        #
+        # Tests with more compact skip connection layers ("block5i_add" (32x32x160),
+        # "block3d_add" (64x64x64), "block2d_add" (128x128x48), "stem_activation"
+        # (256x256x24)) showed that the decrease in model size was small (30 MB) but
+        # the accuracy performance impact was significant (-1.5%).
+        "efficientnetv2b0": (
+            ka.EfficientNetV2B0,
+            ka.efficientnet_v2.preprocess_input,
+            (
+                "block4b_expand_conv",  # 32x32x384
+                "block3a_expand_conv",  # 64x64x128
+                "block2a_expand_activation",  # 128x128x64
+                "stem_activation",  # 256x256x32
+            ),
+        ),
+        "efficientnetv2b1": (
+            ka.EfficientNetV2B1,
+            ka.efficientnet_v2.preprocess_input,
+            (
+                "block4b_expand_conv",  # 32x32x384
+                "block3a_expand_conv",  # 64x64x128
+                "block2a_expand_activation",  # 128x128x64
+                "stem_activation",  # 256x256x32
+            ),
+        ),
+        "efficientnetv2b2": (
+            ka.EfficientNetV2B2,
+            ka.efficientnet_v2.preprocess_input,
+            (
+                "block4b_expand_conv",  # 32x32x416
+                "block3a_expand_conv",  # 64x64x128
+                "block2a_expand_activation",  # 128x128x64
+                "stem_activation",  # 256x256x32
+            ),
+        ),
+        "efficientnetv2b3": (
+            ka.EfficientNetV2B3,
+            ka.efficientnet_v2.preprocess_input,
+            (
+                "block4b_expand_conv",  # 32x32x448
+                "block3a_expand_conv",  # 64x64x160
+                "block2a_expand_activation",  # 128x128x64
+                "stem_activation",  # 256x256x40
+            ),
+        ),
+        "efficientnetv2s": (
+            ka.EfficientNetV2S,
+            ka.efficientnet_v2.preprocess_input,
+            (
+                "block6a_expand_conv",  # 32x32x960
+                "block4a_expand_conv",  # 64x64x256
+                "block2a_expand_activation",  # 128x128x96
+                "stem_activation",  # 256x256x24
+            ),
+        ),
+        "efficientnetv2m": (
+            ka.EfficientNetV2M,
+            ka.efficientnet_v2.preprocess_input,
+            (
+                "block6a_expand_conv",  # 32x32x1056
+                "block4a_expand_conv",  # 64x64x320
+                "block2a_expand_activation",  # 128x128x96
+                "stem_activation",  # 256x256x24
+            ),
+        ),
+        "efficientnetv2l": (
+            ka.EfficientNetV2L,
+            ka.efficientnet_v2.preprocess_input,
+            (
+                "block6a_expand_conv",  # 32x32x1344
+                "block4a_expand_conv",  # 64x64x384
+                "block2a_expand_activation",  # 128x128x128
+                "stem_activation",  # 256x256x32
+            ),
+        ),
+    }
+
+    @property
+    def models(self):
+        """Get the dictionary of available backbones."""
+        return self._models
+
+    def models_names(self):
+        """Get the list of available backbone names."""
+        return list(self.models.keys())
+
+        """Get a backbone model by name."""
+
+    def get_backbone(self, name, *args, **kwargs) -> keras.Model:
+        """Get a backbone model by name."""
+        backbone = self._models.get(name)
+        if backbone is None:
+            raise ValueError(f"Backbone with name '{name}' is not supported.")
+
+        model_fn, _, _ = backbone
+        model = model_fn(*args, **kwargs)
+        return model
+
+    def get_feature_layers(self, name, n=5):
+        """Get the list of skip layers for a backbone by name."""
+        return self._models.get(name)[2][:n]
+
+    def get_preprocessing(self, name):
+        """Get the preprocessing function for a backbone by name."""
+        return self._models.get(name)[1]
+
+    def get_custom_objects(self, name):
+        """Get the custom objects for a backbone by name."""
+        if name == "inceptionresnetv2":
+            return irv2.get_custom_objects()
+        return {}
+
+
+Backbones = BackbonesFactory()

{segmodels_keras-0.1.0.dev11 → segmodels_keras-0.1.0.dev13}/segmodels_keras.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: segmodels_keras
-Version: 0.1.0.dev11
+Version: 0.1.0.dev13
 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.dev11 → segmodels_keras-0.1.0.dev13}/tests/test_models.py

@@ -22,25 +22,25 @@ def str2bool(v):
     return v.lower() in ("yes", "true", "t", "1")
 
 
-def get_test_backbones():
-    is_full = str2bool(os.environ.get("FULL_TEST", "False"))
+def get_test_backbones(with_weights: bool):
+    if with_weights:
+        is_full = str2bool(os.environ.get("FULL_TEST", "False"))
+    else:
+        is_full = str2bool(os.environ.get("FULL_TEST_WEIGHTS", "False"))
+
     if not is_full:
-        return [
-            "resnet34",
-            "resnet50",
-            "inceptionresnetv2",
-            "efficientnetb0",
-            "efficientnetv2s",
-        ]
+        return ["resnet34", "resnet50", "inceptionresnetv2", "efficientnetb0"]
     else:
         return get_available_backbone_names()
 
 
 def keras_test(func):
     """Function wrapper to clean up after TensorFlow tests.
-    # Arguments
+
+    Args:
         func: test function to clean up after.
-    # Returns
+
+    Returns:
         A function wrapping the input function.
     """
 
@@ -53,40 +53,37 @@ def keras_test(func):
     return wrapper
 
 
-@pytest.mark.parametrize("backbone_name", get_test_backbones())
+@pytest.mark.parametrize("backbone_name", get_test_backbones(with_weights=False))
 @pytest.mark.parametrize(
-    "model_name, input_shape, encoder_weights",
+    "model_name, input_shape",
     [
-        ("unet", None, None),
-        ("unet", None, "imagenet"),
-        ("unet", (256, 256, 4), None),
-        ("linknet", None, None),
-        ("linknet", (256, 256, 4), None),
-        ("pspnet", (384, 384, 4), None),
-        ("fpn", None, None),
-        ("fpn", (256, 256, 4), None),
+        ("unet", None),
+        ("unet", (256, 256, 4)),
+        ("linknet", None),
+        ("linknet", (256, 256, 4)),
+        ("pspnet", (384, 384, 4)),
+        ("fpn", None),
+        ("fpn", (256, 256, 4)),
     ],
 )
 @keras_test
-def test_get_model(model_name, backbone_name, input_shape, encoder_weights):
-    """Test all segmentation models with different backbones.
+def test_get_model(model_name, backbone_name, input_shape):
+    """Test creation of model/backbone combinations.
 
-    input_shape=None means any shape (32x32 used in test), otherwise a fixed shape is
-    used.
+    No weights are used in this test to avoid downloading pretrained models.
     """
-    n_channels = 3 if input_shape is None else input_shape[-1]
-    test_shape = (1, 32, 32, n_channels) if input_shape is None else (1, *input_shape)
+    _assert_get_model_output_shape(model_name, backbone_name, input_shape, None)
 
-    x = np.ones(test_shape)
-    model = get_model(
-        model_name,
-        backbone_name=backbone_name,
-        input_shape=input_shape or (None, None, n_channels),
-        encoder_weights=encoder_weights,
-    )
-    y = model.predict(x)
 
-    assert x.shape[:-1] == y.shape[:-1]
+@pytest.mark.parametrize("backbone_name", get_test_backbones(with_weights=True))
+@keras_test
+def test_get_model_with_imagenet_weights(backbone_name):
+    """Test model creation with ImageNet-pretrained encoder weights.
+
+    Only tested on a limited set of backbones as this will download the pretrained
+    weights.
+    """
+    _assert_get_model_output_shape("unet", backbone_name, None, "imagenet")
 
 
 def test_get_model_invalid_name():
@@ -113,5 +110,19 @@ def test_get_model_weights_notop_keras3():
     assert len(model.layers) > 0
 
 
-if __name__ == "__main__":
-    pytest.main([__file__])
+def _assert_get_model_output_shape(
+    model_name, backbone_name, input_shape, encoder_weights
+):
+    n_channels = 3 if input_shape is None else input_shape[-1]
+    test_shape = (1, 32, 32, n_channels) if input_shape is None else (1, *input_shape)
+
+    x = np.ones(test_shape)
+    model = get_model(
+        model_name,
+        backbone_name=backbone_name,
+        input_shape=input_shape or (None, None, n_channels),
+        encoder_weights=encoder_weights,
+    )
+    y = model.predict(x)
+
+    assert x.shape[:-1] == y.shape[:-1]

segmodels_keras-0.1.0.dev11/segmodels_keras/_version.py

@@ -1,2 +0,0 @@
-__versioninfo__: tuple[int, int, int, str] = (0, 1, 0, "dev11")
-__version__: str = ".".join(map(str, __versioninfo__))

segmodels_keras-0.1.0.dev11/segmodels_keras/backbones/backbones_factory.py

@@ -1,326 +0,0 @@
-"""Backbones factory module."""
-
-import keras
-import keras.applications as ka
-
-from . import inception_resnet_v2 as irv2
-from . import inception_v3 as iv3
-from . import resnet_18_34
-
-
-class BackbonesFactory:
-    """Dict with all supported backbones.
-
-    Each backbone is represented as a tuple of 3 elements:
-    1. Backbone class
-    2. Preprocessing function
-    3. List of layers to take features from backbone in the following order:
-       (x16, x8, x4, x2, x1) - `x4` mean that features has 4 times less spatial
-       resolution (Height x Width) than input image.
-    """
-
-    _models = {  # noqa: RUF012
-        # ResNets < 50 layers are NOT available via keras.applications, so use a
-        # torchvision-compatible post-activation BasicBlock implementation.
-        # Layer names follow the keras.applications ResNet50+ naming convention.
-        "resnet18": (
-            resnet_18_34.ResNet18,
-            resnet_18_34.preprocess_input,
-            # resnet18: layer1=2, layer2=2, layer3=2, layer4=2 blocks
-            # backbone output = conv5_block2_out (512ch); skips are the 4 stages before
-            ("conv4_block2_out", "conv3_block2_out", "conv2_block2_out", "conv1_relu"),
-        ),
-        "resnet34": (
-            resnet_18_34.ResNet34,
-            resnet_18_34.preprocess_input,
-            # resnet34: layer1=3, layer2=4, layer3=6, layer4=3 blocks
-            # backbone output = conv5_block3_out (512ch); skips are the 4 stages before
-            ("conv4_block6_out", "conv3_block4_out", "conv2_block3_out", "conv1_relu"),
-        ),
-        # ResNets > 50 layers are available via keras.applications, so use those.
-        # Skip layers (inverted) from https://github.com/yingkaisha/keras-unet-collection
-        "resnet50": (
-            ka.ResNet50,
-            ka.resnet.preprocess_input,
-            ("conv4_block6_out", "conv3_block4_out", "conv2_block3_out", "conv1_relu"),
-        ),
-        "resnet101": (
-            ka.ResNet101,
-            ka.resnet.preprocess_input,
-            ("conv4_block23_out", "conv3_block4_out", "conv2_block3_out", "conv1_relu"),
-        ),
-        "resnet152": (
-            ka.ResNet152,
-            ka.resnet.preprocess_input,
-            ("conv4_block36_out", "conv3_block8_out", "conv2_block3_out", "conv1_relu"),
-        ),
-        # ResNetV2
-        # Skip layers (inverted) from https://github.com/yingkaisha/keras-unet-collection
-        "resnet50v2": (
-            ka.ResNet50V2,
-            ka.resnet_v2.preprocess_input,
-            (
-                "conv4_block6_1_relu",
-                "conv3_block4_1_relu",
-                "conv2_block3_1_relu",
-                "conv1_conv",
-            ),
-        ),
-        "resnet101v2": (
-            ka.ResNet101V2,
-            ka.resnet_v2.preprocess_input,
-            (
-                "conv4_block23_1_relu",
-                "conv3_block4_1_relu",
-                "conv2_block3_1_relu",
-                "conv1_conv",
-            ),
-        ),
-        "resnet152v2": (
-            ka.ResNet152V2,
-            ka.resnet_v2.preprocess_input,
-            (
-                "conv4_block36_1_relu",
-                "conv3_block8_1_relu",
-                "conv2_block3_1_relu",
-                "conv1_conv",
-            ),
-        ),
-        # VGG
-        # Skip layers from segmentation_models
-        "vgg16": (
-            ka.vgg16.VGG16,
-            ka.vgg16.preprocess_input,
-            (
-                "block5_conv3",
-                "block4_conv3",
-                "block3_conv3",
-                "block2_conv2",
-                "block1_conv2",
-            ),
-        ),
-        "vgg19": (
-            ka.vgg19.VGG19,
-            ka.vgg19.preprocess_input,
-            (
-                "block5_conv4",
-                "block4_conv4",
-                "block3_conv4",
-                "block2_conv2",
-                "block1_conv2",
-            ),
-        ),
-        # DenseNet
-        # Skip layers (inverted) from https://github.com/yingkaisha/keras-unet-collection
-        "densenet121": (
-            ka.densenet.DenseNet121,
-            ka.densenet.preprocess_input,
-            # (311, 139, 51, 4),
-            ("pool4_conv", "pool3_conv", "pool2_conv", "conv1/relu"),
-        ),
-        "densenet169": (
-            ka.densenet.DenseNet169,
-            ka.densenet.preprocess_input,
-            # (367, 139, 51, 4),
-            ("pool4_conv", "pool3_conv", "pool2_conv", "conv1/relu"),
-        ),
-        "densenet201": (
-            ka.densenet.DenseNet201,
-            ka.densenet.preprocess_input,
-            # (479, 139, 51, 4),
-            ("pool4_conv", "pool3_conv", "pool2_conv", "conv1/relu"),
-        ),
-        # Inception
-        # Skip layers from segmentation_models
-        "inceptionresnetv2": (
-            irv2.InceptionResNetV2,
-            irv2.preprocess_input,
-            # 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
-        # Skip layers from segmentation_models
-        "mobilenet": (
-            ka.mobilenet.MobileNet,
-            ka.mobilenet.preprocess_input,
-            (
-                "conv_pw_11_relu",
-                "conv_pw_5_relu",
-                "conv_pw_3_relu",
-                "conv_pw_1_relu",
-            ),
-        ),
-        "mobilenetv2": (
-            ka.mobilenet_v2.MobileNetV2,
-            ka.mobilenet_v2.preprocess_input,
-            (
-                "block_13_expand_relu",
-                "block_6_expand_relu",
-                "block_3_expand_relu",
-                "block_1_expand_relu",
-            ),
-        ),
-        # EfficientNet
-        # Skip layers from segmentation_models
-        "efficientnetb0": [
-            ka.EfficientNetB0,
-            ka.efficientnet.preprocess_input,
-            (
-                "block6a_expand_activation",
-                "block4a_expand_activation",
-                "block3a_expand_activation",
-                "block2a_expand_activation",
-            ),
-        ],
-        "efficientnetb1": [
-            ka.EfficientNetB1,
-            ka.efficientnet.preprocess_input,
-            (
-                "block6a_expand_activation",
-                "block4a_expand_activation",
-                "block3a_expand_activation",
-                "block2a_expand_activation",
-            ),
-        ],
-        "efficientnetb2": [
-            ka.EfficientNetB2,
-            ka.efficientnet.preprocess_input,
-            (
-                "block6a_expand_activation",
-                "block4a_expand_activation",
-                "block3a_expand_activation",
-                "block2a_expand_activation",
-            ),
-        ],
-        "efficientnetb3": [
-            ka.EfficientNetB3,
-            ka.efficientnet.preprocess_input,
-            (
-                "block6a_expand_activation",
-                "block4a_expand_activation",
-                "block3a_expand_activation",
-                "block2a_expand_activation",
-            ),
-        ],
-        "efficientnetb4": [
-            ka.EfficientNetB4,
-            ka.efficientnet.preprocess_input,
-            (
-                "block6a_expand_activation",
-                "block4a_expand_activation",
-                "block3a_expand_activation",
-                "block2a_expand_activation",
-            ),
-        ],
-        "efficientnetb5": [
-            ka.EfficientNetB5,
-            ka.efficientnet.preprocess_input,
-            (
-                "block6a_expand_activation",
-                "block4a_expand_activation",
-                "block3a_expand_activation",
-                "block2a_expand_activation",
-            ),
-        ],
-        "efficientnetb6": [
-            ka.EfficientNetB6,
-            ka.efficientnet.preprocess_input,
-            (
-                "block6a_expand_activation",
-                "block4a_expand_activation",
-                "block3a_expand_activation",
-                "block2a_expand_activation",
-            ),
-        ],
-        "efficientnetb7": [
-            ka.EfficientNetB7,
-            ka.efficientnet.preprocess_input,
-            (
-                "block6a_expand_activation",
-                "block4a_expand_activation",
-                "block3a_expand_activation",
-                "block2a_expand_activation",
-            ),
-        ],
-        # EfficientNetV2
-        # Stage layout differs from EfficientNet-Bx.
-        # Skip layers selected to match decoder resolutions:
-        # x16 -> x8 -> x4 -> x2
-        "efficientnetv2s": (
-            ka.EfficientNetV2S,
-            ka.efficientnet_v2.preprocess_input,
-            (
-                "block6a_expand_conv",  # 14x14
-                "block4a_expand_conv",  # 28x28
-                "block2a_expand_activation",  # 56x56
-                "stem_activation",  # 112x112
-            ),
-        ),
-        "efficientnetv2m": (
-            ka.EfficientNetV2M,
-            ka.efficientnet_v2.preprocess_input,
-            (
-                "block6a_expand_conv",
-                "block4a_expand_conv",
-                "block2a_expand_activation",
-                "stem_activation",
-            ),
-        ),
-        "efficientnetv2l": (
-            ka.EfficientNetV2L,
-            ka.efficientnet_v2.preprocess_input,
-            (
-                "block6a_expand_conv",
-                "block4a_expand_conv",
-                "block2a_expand_activation",
-                "stem_activation",
-            ),
-        ),
-    }
-
-    @property
-    def models(self):
-        """Get the dictionary of available backbones."""
-        return self._models
-
-    def models_names(self):
-        """Get the list of available backbone names."""
-        return list(self.models.keys())
-
-        """Get a backbone model by name."""
-
-    def get_backbone(self, name, *args, **kwargs) -> keras.Model:
-        """Get a backbone model by name."""
-        backbone = self._models.get(name)
-        if backbone is None:
-            raise ValueError(f"Backbone with name '{name}' is not supported.")
-
-        model_fn, _, _ = backbone
-        model = model_fn(*args, **kwargs)
-        return model
-
-    def get_feature_layers(self, name, n=5):
-        """Get the list of skip layers for a backbone by name."""
-        return self._models.get(name)[2][:n]
-
-    def get_preprocessing(self, name):
-        """Get the preprocessing function for a backbone by name."""
-        return self._models.get(name)[1]
-
-    def get_custom_objects(self, name):
-        """Get the custom objects for a backbone by name."""
-        if name == "inceptionresnetv2":
-            return irv2.get_custom_objects()
-        return {}
-
-
-Backbones = BackbonesFactory()