Bump segmentation-models-pytorch from 0.4.0 to 0.5.0 in /requirements (#2740)

Author: dependabot[bot]

pyproject.toml

@@ -65,8 +65,8 @@ dependencies = [
     "rasterio>=1.3.3,!=1.4.0,!=1.4.1,!=1.4.2",
     # rtree 1.0.1+ required for Python 3.11 wheels
     "rtree>=1.0.1",
-    # segmentation-models-pytorch 0.3.3+ required for timm 0.8+ support
-    "segmentation-models-pytorch>=0.3.3",
+    # segmentation-models-pytorch 0.5+ required for new UnetDecoder API
+    "segmentation-models-pytorch>=0.5",
     # shapely 1.8.5+ required for Python 3.11 wheels
     "shapely>=1.8.5",
     # timm 0.8+ required for timm.models.adapt_input_conv, 0.9.2 required by SMP

requirements/min-reqs.old

@@ -14,7 +14,7 @@ pillow==9.2.0
 pyproj==3.4.0
 rasterio==1.3.11
 rtree==1.0.1
-segmentation-models-pytorch==0.3.3
+segmentation-models-pytorch==0.5.0
 shapely==1.8.5
 timm==0.9.2
 torch==2.0.0

requirements/required.txt

@@ -14,7 +14,7 @@ pillow==11.2.1
 pyproj==3.7.1
 rasterio==1.4.3
 rtree==1.4.0
-segmentation-models-pytorch==0.4.0
+segmentation-models-pytorch==0.5.0
 shapely==2.0.7
 timm==1.0.15
 torch==2.6.0

torchgeo/models/fcsiam.py

@@ -26,7 +26,7 @@ def __init__(
         encoder_name: str = 'resnet34',
         encoder_depth: int = 5,
         encoder_weights: str | None = 'imagenet',
-        decoder_use_batchnorm: bool = True,
+        decoder_use_batchnorm: bool | str | dict[str, Any] = 'batchnorm',
         decoder_channels: Sequence[int] = (256, 128, 64, 32, 16),
         decoder_attention_type: str | None = None,
         in_channels: int = 3,
@@ -50,10 +50,26 @@ def __init__(
             decoder_channels: List of integers which specify **in_channels**
                 parameter for convolutions used in decoder. Length of the list
                 should be the same as **encoder_depth**
-            decoder_use_batchnorm: If **True**, BatchNorm2d layer between
-                Conv2D and Activation layers is used. If **"inplace"** InplaceABN
-                will be used, allows to decrease memory consumption. Available
-                options are **True, False, "inplace"**
+            decoder_use_batchnorm: Specifies normalization between Conv2D and
+                activation. Accepts the following types:
+
+                - **True**: Defaults to `"batchnorm"`.
+                - **False**: No normalization (`nn.Identity`).
+                - **str**: Specifies normalization type using default parameters.
+                  Available values: `"batchnorm"`, `"identity"`, `"layernorm"`,
+                  `"instancenorm"`, `"inplace"`.
+                - **dict**: Fully customizable normalization settings. Structure:
+                  ```python
+                  {"type": <norm_type>, **kwargs}
+                  ```
+                  where `norm_name` corresponds to normalization type (see above), and
+                  `kwargs` are passed directly to the normalization layer as defined in
+                  PyTorch documentation.
+
+                  **Example**:
+                  ```python
+                  decoder_use_norm={"type": "layernorm", "eps": 1e-2}
+                  ```
             decoder_attention_type: Attention module used in decoder of the model.
                 Available options are **None** and **scse**. SCSE paper
                 https://arxiv.org/abs/1808.08127
@@ -79,9 +95,9 @@ def __init__(
             encoder_channels=encoder_out_channels,
             decoder_channels=decoder_channels,
             n_blocks=encoder_depth,
-            use_batchnorm=decoder_use_batchnorm,
-            center=True if encoder_name.startswith('vgg') else False,
+            use_norm=decoder_use_batchnorm,
             attention_type=decoder_attention_type,
+            add_center_block=True if encoder_name.startswith('vgg') else False,
         )
 
         self.segmentation_head = smp.base.SegmentationHead(
@@ -111,7 +127,7 @@ def forward(self, x: Tensor) -> Tensor:
             for i in range(1, len(features1))
         ]
         features.insert(0, features2[0])
-        decoder_output = self.decoder(*features)
+        decoder_output = self.decoder(features)
         masks: Tensor = self.segmentation_head(decoder_output)
         return masks
 
@@ -150,6 +166,6 @@ def forward(self, x: Tensor) -> Tensor:
         features1, features2 = self.encoder(x1), self.encoder(x2)
         features = [features2[i] - features1[i] for i in range(1, len(features1))]
         features.insert(0, features2[0])
-        decoder_output = self.decoder(*features)
+        decoder_output = self.decoder(features)
         masks: Tensor = self.segmentation_head(decoder_output)
         return masks