Expose timm constructor arguments (#960)

* Expose timm constructor arguments

* Remove leak from other branch

* Rename dupls to duplicates

Author: DimitrisMantas

segmentation_models_pytorch/decoders/deeplabv3/model.py

@@ -1,11 +1,12 @@
-from typing import Optional
+from typing import Any, Optional
 
 from segmentation_models_pytorch.base import (
-    SegmentationModel,
-    SegmentationHead,
     ClassificationHead,
+    SegmentationHead,
+    SegmentationModel,
 )
 from segmentation_models_pytorch.encoders import get_encoder
+
 from .decoder import DeepLabV3Decoder, DeepLabV3PlusDecoder
 
 
@@ -36,6 +37,8 @@ class DeepLabV3(SegmentationModel):
                 - dropout (float): Dropout factor in [0, 1)
                 - activation (str): An activation function to apply "sigmoid"/"softmax"
                     (could be **None** to return logits)
+        kwargs: Arguments passed to the encoder class ``__init__()`` function. Applies only to ``timm`` models. Keys with ``None`` values are pruned before passing.
+
     Returns:
         ``torch.nn.Module``: **DeepLabV3**
 
@@ -55,6 +58,7 @@ def __init__(
         activation: Optional[str] = None,
         upsampling: int = 8,
         aux_params: Optional[dict] = None,
+        **kwargs: dict[str, Any],
     ):
         super().__init__()
 
@@ -64,6 +68,7 @@ def __init__(
             depth=encoder_depth,
             weights=encoder_weights,
             output_stride=8,
+            **kwargs,
         )
 
         self.decoder = DeepLabV3Decoder(
@@ -116,6 +121,8 @@ class DeepLabV3Plus(SegmentationModel):
                 - dropout (float): Dropout factor in [0, 1)
                 - activation (str): An activation function to apply "sigmoid"/"softmax"
                     (could be **None** to return logits)
+        kwargs: Arguments passed to the encoder class ``__init__()`` function. Applies only to ``timm`` models. Keys with ``None`` values are pruned before passing.
+
     Returns:
         ``torch.nn.Module``: **DeepLabV3Plus**
 
@@ -137,6 +144,7 @@ def __init__(
         activation: Optional[str] = None,
         upsampling: int = 4,
         aux_params: Optional[dict] = None,
+        **kwargs: dict[str, Any],
     ):
         super().__init__()
 
@@ -153,6 +161,7 @@ def __init__(
             depth=encoder_depth,
             weights=encoder_weights,
             output_stride=encoder_output_stride,
+            **kwargs,
         )
 
         self.decoder = DeepLabV3PlusDecoder(

segmentation_models_pytorch/decoders/fpn/model.py

@@ -1,11 +1,12 @@
-from typing import Optional
+from typing import Any, Optional
 
 from segmentation_models_pytorch.base import (
-    SegmentationModel,
-    SegmentationHead,
     ClassificationHead,
+    SegmentationHead,
+    SegmentationModel,
 )
 from segmentation_models_pytorch.encoders import get_encoder
+
 from .decoder import FPNDecoder
 
 
@@ -40,6 +41,7 @@ class FPN(SegmentationModel):
                 - dropout (float): Dropout factor in [0, 1)
                 - activation (str): An activation function to apply "sigmoid"/"softmax"
                     (could be **None** to return logits)
+        kwargs: Arguments passed to the encoder class ``__init__()`` function. Applies only to ``timm`` models. Keys with ``None`` values are pruned before passing.
 
     Returns:
         ``torch.nn.Module``: **FPN**
@@ -63,6 +65,7 @@ def __init__(
         activation: Optional[str] = None,
         upsampling: int = 4,
         aux_params: Optional[dict] = None,
+        **kwargs: dict[str, Any],
     ):
         super().__init__()
 
@@ -77,6 +80,7 @@ def __init__(
             in_channels=in_channels,
             depth=encoder_depth,
             weights=encoder_weights,
+            **kwargs,
         )
 
         self.decoder = FPNDecoder(

segmentation_models_pytorch/decoders/linknet/model.py

@@ -1,11 +1,12 @@
-from typing import Optional, Union
+from typing import Any, Optional, Union
 
 from segmentation_models_pytorch.base import (
+    ClassificationHead,
     SegmentationHead,
     SegmentationModel,
-    ClassificationHead,
 )
 from segmentation_models_pytorch.encoders import get_encoder
+
 from .decoder import LinknetDecoder
 
 
@@ -43,6 +44,7 @@ class Linknet(SegmentationModel):
                 - dropout (float): Dropout factor in [0, 1)
                 - activation (str): An activation function to apply "sigmoid"/"softmax"
                     (could be **None** to return logits)
+        kwargs: Arguments passed to the encoder class ``__init__()`` function. Applies only to ``timm`` models. Keys with ``None`` values are pruned before passing.
 
     Returns:
         ``torch.nn.Module``: **Linknet**
@@ -61,6 +63,7 @@ def __init__(
         classes: int = 1,
         activation: Optional[Union[str, callable]] = None,
         aux_params: Optional[dict] = None,
+        **kwargs: dict[str, Any],
     ):
         super().__init__()
 
@@ -74,6 +77,7 @@ def __init__(
             in_channels=in_channels,
             depth=encoder_depth,
             weights=encoder_weights,
+            **kwargs,
         )
 
         self.decoder = LinknetDecoder(

segmentation_models_pytorch/decoders/manet/model.py

@@ -1,11 +1,12 @@
-from typing import Optional, Union, List
+from typing import Any, List, Optional, Union
 
-from segmentation_models_pytorch.encoders import get_encoder
 from segmentation_models_pytorch.base import (
-    SegmentationModel,
-    SegmentationHead,
     ClassificationHead,
+    SegmentationHead,
+    SegmentationModel,
 )
+from segmentation_models_pytorch.encoders import get_encoder
+
 from .decoder import MAnetDecoder
 
 
@@ -45,6 +46,7 @@ class MAnet(SegmentationModel):
                 - dropout (float): Dropout factor in [0, 1)
                 - activation (str): An activation function to apply "sigmoid"/"softmax"
                     (could be **None** to return logits)
+        kwargs: Arguments passed to the encoder class ``__init__()`` function. Applies only to ``timm`` models. Keys with ``None`` values are pruned before passing.
 
     Returns:
         ``torch.nn.Module``: **MAnet**
@@ -66,6 +68,7 @@ def __init__(
         classes: int = 1,
         activation: Optional[Union[str, callable]] = None,
         aux_params: Optional[dict] = None,
+        **kwargs: dict[str, Any],
     ):
         super().__init__()
 
@@ -74,6 +77,7 @@ def __init__(
             in_channels=in_channels,
             depth=encoder_depth,
             weights=encoder_weights,
+            **kwargs,
         )
 
         self.decoder = MAnetDecoder(

segmentation_models_pytorch/decoders/pan/model.py

@@ -1,11 +1,12 @@
-from typing import Optional, Union
+from typing import Any, Optional, Union
 
-from segmentation_models_pytorch.encoders import get_encoder
 from segmentation_models_pytorch.base import (
-    SegmentationModel,
-    SegmentationHead,
     ClassificationHead,
+    SegmentationHead,
+    SegmentationModel,
 )
+from segmentation_models_pytorch.encoders import get_encoder
+
 from .decoder import PANDecoder
 
 
@@ -38,6 +39,7 @@ class PAN(SegmentationModel):
                 - dropout (float): Dropout factor in [0, 1)
                 - activation (str): An activation function to apply "sigmoid"/"softmax"
                     (could be **None** to return logits)
+        kwargs: Arguments passed to the encoder class ``__init__()`` function. Applies only to ``timm`` models. Keys with ``None`` values are pruned before passing.
 
     Returns:
         ``torch.nn.Module``: **PAN**
@@ -58,6 +60,7 @@ def __init__(
         activation: Optional[Union[str, callable]] = None,
         upsampling: int = 4,
         aux_params: Optional[dict] = None,
+        **kwargs: dict[str, Any],
     ):
         super().__init__()
 
@@ -74,6 +77,7 @@ def __init__(
             depth=5,
             weights=encoder_weights,
             output_stride=encoder_output_stride,
+            **kwargs,
         )
 
         self.decoder = PANDecoder(

segmentation_models_pytorch/decoders/pspnet/model.py

@@ -1,11 +1,12 @@
-from typing import Optional, Union
+from typing import Any, Optional, Union
 
-from segmentation_models_pytorch.encoders import get_encoder
 from segmentation_models_pytorch.base import (
-    SegmentationModel,
-    SegmentationHead,
     ClassificationHead,
+    SegmentationHead,
+    SegmentationModel,
 )
+from segmentation_models_pytorch.encoders import get_encoder
+
 from .decoder import PSPDecoder
 
 
@@ -44,6 +45,7 @@ class PSPNet(SegmentationModel):
                 - dropout (float): Dropout factor in [0, 1)
                 - activation (str): An activation function to apply "sigmoid"/"softmax"
                     (could be **None** to return logits)
+        kwargs: Arguments passed to the encoder class ``__init__()`` function. Applies only to ``timm`` models. Keys with ``None`` values are pruned before passing.
 
     Returns:
         ``torch.nn.Module``: **PSPNet**
@@ -65,6 +67,7 @@ def __init__(
         activation: Optional[Union[str, callable]] = None,
         upsampling: int = 8,
         aux_params: Optional[dict] = None,
+        **kwargs: dict[str, Any],
     ):
         super().__init__()
 
@@ -73,6 +76,7 @@ def __init__(
             in_channels=in_channels,
             depth=encoder_depth,
             weights=encoder_weights,
+            **kwargs,
         )
 
         self.decoder = PSPDecoder(

segmentation_models_pytorch/decoders/unet/model.py

@@ -1,11 +1,12 @@
-from typing import Optional, Union, List
+from typing import Any, List, Optional, Union
 
-from segmentation_models_pytorch.encoders import get_encoder
 from segmentation_models_pytorch.base import (
-    SegmentationModel,
-    SegmentationHead,
     ClassificationHead,
+    SegmentationHead,
+    SegmentationModel,
 )
+from segmentation_models_pytorch.encoders import get_encoder
+
 from .decoder import UnetDecoder
 
 
@@ -44,6 +45,7 @@ class Unet(SegmentationModel):
                 - dropout (float): Dropout factor in [0, 1)
                 - activation (str): An activation function to apply "sigmoid"/"softmax"
                     (could be **None** to return logits)
+        kwargs: Arguments passed to the encoder class ``__init__()`` function. Applies only to ``timm`` models. Keys with ``None`` values are pruned before passing.
 
     Returns:
         ``torch.nn.Module``: Unet
@@ -65,6 +67,7 @@ def __init__(
         classes: int = 1,
         activation: Optional[Union[str, callable]] = None,
         aux_params: Optional[dict] = None,
+        **kwargs: dict[str, Any],
     ):
         super().__init__()
 
@@ -73,6 +76,7 @@ def __init__(
             in_channels=in_channels,
             depth=encoder_depth,
             weights=encoder_weights,
+            **kwargs,
         )
 
         self.decoder = UnetDecoder(

segmentation_models_pytorch/decoders/unetplusplus/model.py

@@ -1,11 +1,12 @@
-from typing import Optional, Union, List
+from typing import Any, List, Optional, Union
 
-from segmentation_models_pytorch.encoders import get_encoder
 from segmentation_models_pytorch.base import (
-    SegmentationModel,
-    SegmentationHead,
     ClassificationHead,
+    SegmentationHead,
+    SegmentationModel,
 )
+from segmentation_models_pytorch.encoders import get_encoder
+
 from .decoder import UnetPlusPlusDecoder
 
 
@@ -44,6 +45,7 @@ class UnetPlusPlus(SegmentationModel):
                 - dropout (float): Dropout factor in [0, 1)
                 - activation (str): An activation function to apply "sigmoid"/"softmax"
                     (could be **None** to return logits)
+        kwargs: Arguments passed to the encoder class ``__init__()`` function. Applies only to ``timm`` models. Keys with ``None`` values are pruned before passing.
 
     Returns:
         ``torch.nn.Module``: **Unet++**
@@ -65,6 +67,7 @@ def __init__(
         classes: int = 1,
         activation: Optional[Union[str, callable]] = None,
         aux_params: Optional[dict] = None,
+        **kwargs: dict[str, Any],
     ):
         super().__init__()
 
@@ -78,6 +81,7 @@ def __init__(
             in_channels=in_channels,
             depth=encoder_depth,
             weights=encoder_weights,
+            **kwargs,
         )
 
         self.decoder = UnetPlusPlusDecoder(

segmentation_models_pytorch/decoders/upernet/model.py

@@ -1,11 +1,12 @@
-from typing import Optional, Union
+from typing import Any, Optional, Union
 
-from segmentation_models_pytorch.encoders import get_encoder
 from segmentation_models_pytorch.base import (
-    SegmentationModel,
-    SegmentationHead,
     ClassificationHead,
+    SegmentationHead,
+    SegmentationModel,
 )
+from segmentation_models_pytorch.encoders import get_encoder
+
 from .decoder import UPerNetDecoder
 
 
@@ -36,6 +37,7 @@ class UPerNet(SegmentationModel):
                 - dropout (float): Dropout factor in [0, 1)
                 - activation (str): An activation function to apply "sigmoid"/"softmax"
                     (could be **None** to return logits)
+        kwargs: Arguments passed to the encoder class ``__init__()`` function. Applies only to ``timm`` models. Keys with ``None`` values are pruned before passing.
 
     Returns:
         ``torch.nn.Module``: **UPerNet**
@@ -56,6 +58,7 @@ def __init__(
         classes: int = 1,
         activation: Optional[Union[str, callable]] = None,
         aux_params: Optional[dict] = None,
+        **kwargs: dict[str, Any],
     ):
         super().__init__()
 
@@ -64,6 +67,7 @@ def __init__(
             in_channels=in_channels,
             depth=encoder_depth,
             weights=encoder_weights,
+            **kwargs,
         )
 
         self.decoder = UPerNetDecoder(