From 045593671ee7bbcb3db107d3f69468f21ea43c48 Mon Sep 17 00:00:00 2001
From: Ryan <23580140+brianhou0208@users.noreply.github.com>
Date: Sat, 30 Nov 2024 02:11:36 +0800
Subject: [PATCH 1/7] fix encoder depth & output stride

---
 .../decoders/deeplabv3/decoder.py             | 31 ++++++++--------
 .../decoders/deeplabv3/model.py               | 37 ++++++++++++++++---
 2 files changed, 46 insertions(+), 22 deletions(-)

diff --git a/segmentation_models_pytorch/decoders/deeplabv3/decoder.py b/segmentation_models_pytorch/decoders/deeplabv3/decoder.py
index caeb95d1..8f2668bc 100644
--- a/segmentation_models_pytorch/decoders/deeplabv3/decoder.py
+++ b/segmentation_models_pytorch/decoders/deeplabv3/decoder.py
@@ -61,7 +61,6 @@ def __init__(
             nn.BatchNorm2d(out_channels),
             nn.ReLU(),
         )
-        self.out_channels = out_channels
 
     def forward(self, *features):
         return super().forward(features[-1])
@@ -71,6 +70,7 @@ class DeepLabV3PlusDecoder(nn.Module):
     def __init__(
         self,
         encoder_channels: Sequence[int, ...],
+        encoder_depth: int,
         out_channels: int,
         atrous_rates: Iterable[int],
         output_stride: Literal[8, 16],
@@ -78,14 +78,13 @@ def __init__(
         aspp_dropout: float,
     ):
         super().__init__()
-        if output_stride not in {8, 16}:
+        if encoder_depth < 3:
             raise ValueError(
-                "Output stride should be 8 or 16, got {}.".format(output_stride)
+                "Encoder depth for DeepLabV3Plus decoder cannot be less than 3, got {}.".format(
+                    encoder_depth
+                )
             )
 
-        self.out_channels = out_channels
-        self.output_stride = output_stride
-
         self.aspp = nn.Sequential(
             ASPP(
                 encoder_channels[-1],
@@ -101,10 +100,10 @@ def __init__(
             nn.ReLU(),
         )
 
-        scale_factor = 2 if output_stride == 8 else 4
+        scale_factor = 4 if output_stride == 16 and encoder_depth > 3 else 2
         self.up = nn.UpsamplingBilinear2d(scale_factor=scale_factor)
 
-        highres_in_channels = encoder_channels[-4]
+        highres_in_channels = encoder_channels[2]
         highres_out_channels = 48  # proposed by authors of paper
         self.block1 = nn.Sequential(
             nn.Conv2d(
@@ -128,7 +127,7 @@ def __init__(
     def forward(self, *features):
         aspp_features = self.aspp(features[-1])
         aspp_features = self.up(aspp_features)
-        high_res_features = self.block1(features[-4])
+        high_res_features = self.block1(features[2])
         concat_features = torch.cat([aspp_features, high_res_features], dim=1)
         fused_features = self.block2(concat_features)
         return fused_features
@@ -228,13 +227,13 @@ def forward(self, x):
 class SeparableConv2d(nn.Sequential):
     def __init__(
         self,
-        in_channels,
-        out_channels,
-        kernel_size,
-        stride=1,
-        padding=0,
-        dilation=1,
-        bias=True,
+        in_channels: int,
+        out_channels: int,
+        kernel_size: int,
+        stride: int = 1,
+        padding: int = 0,
+        dilation: int = 1,
+        bias: bool = True,
     ):
         dephtwise_conv = nn.Conv2d(
             in_channels,
diff --git a/segmentation_models_pytorch/decoders/deeplabv3/model.py b/segmentation_models_pytorch/decoders/deeplabv3/model.py
index d67a3be3..030e42fd 100644
--- a/segmentation_models_pytorch/decoders/deeplabv3/model.py
+++ b/segmentation_models_pytorch/decoders/deeplabv3/model.py
@@ -35,7 +35,7 @@ class DeepLabV3(SegmentationModel):
             Available options are **"sigmoid"**, **"softmax"**, **"logsoftmax"**, **"tanh"**, **"identity"**,
                 **callable** and **None**.
             Default is **None**
-        upsampling: Final upsampling factor (should have the same value as ``encoder_output_stride`` to preserve input-output spatial shape identity).
+        upsampling: Final upsampling factor. Default is **None** to preserve input-output spatial shape identity
         aux_params: Dictionary with parameters of the auxiliary output (classification head). Auxiliary output is build
             on top of encoder if **aux_params** is not **None** (default). Supported params:
                 - classes (int): A number of classes
@@ -43,7 +43,8 @@ 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.
+        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**
@@ -72,6 +73,13 @@ def __init__(
     ):
         super().__init__()
 
+        if encoder_output_stride not in [8, 16]:
+            raise ValueError(
+                "DeeplabV3 support output stride 8 or 16, got {}.".format(
+                    encoder_output_stride
+                )
+            )
+        
         self.encoder = get_encoder(
             encoder_name,
             in_channels=in_channels,
@@ -81,6 +89,14 @@ def __init__(
             **kwargs,
         )
 
+        if upsampling is None:
+            if encoder_depth <= 3:
+                scale_factor = 2 ** encoder_depth
+            else:
+                scale_factor = encoder_output_stride
+        else:
+            scale_factor = upsampling
+
         self.decoder = DeepLabV3Decoder(
             in_channels=self.encoder.out_channels[-1],
             out_channels=decoder_channels,
@@ -90,11 +106,11 @@ def __init__(
         )
 
         self.segmentation_head = SegmentationHead(
-            in_channels=self.decoder.out_channels,
+            in_channels=decoder_channels,
             out_channels=classes,
             activation=activation,
             kernel_size=1,
-            upsampling=encoder_output_stride if upsampling is None else upsampling,
+            upsampling=scale_factor,
         )
 
         if aux_params is not None:
@@ -137,7 +153,8 @@ 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.
+        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**
@@ -166,6 +183,13 @@ def __init__(
     ):
         super().__init__()
 
+        if encoder_output_stride not in [8, 16]:
+            raise ValueError(
+                "DeeplabV3Plus support output stride 8 or 16, got {}.".format(
+                    encoder_output_stride
+                )
+            )
+
         self.encoder = get_encoder(
             encoder_name,
             in_channels=in_channels,
@@ -177,6 +201,7 @@ def __init__(
 
         self.decoder = DeepLabV3PlusDecoder(
             encoder_channels=self.encoder.out_channels,
+            encoder_depth=encoder_depth,
             out_channels=decoder_channels,
             atrous_rates=decoder_atrous_rates,
             output_stride=encoder_output_stride,
@@ -185,7 +210,7 @@ def __init__(
         )
 
         self.segmentation_head = SegmentationHead(
-            in_channels=self.decoder.out_channels,
+            in_channels=decoder_channels,
             out_channels=classes,
             activation=activation,
             kernel_size=1,

From 79d55685934e9d7496dea9a49d966b13165a0cf8 Mon Sep 17 00:00:00 2001
From: Ryan <23580140+brianhou0208@users.noreply.github.com>
Date: Sat, 30 Nov 2024 02:16:23 +0800
Subject: [PATCH 2/7] fix ruff style

---
 segmentation_models_pytorch/__init__.py                 | 4 +++-
 segmentation_models_pytorch/decoders/deeplabv3/model.py | 8 ++++----
 2 files changed, 7 insertions(+), 5 deletions(-)

diff --git a/segmentation_models_pytorch/__init__.py b/segmentation_models_pytorch/__init__.py
index c8209226..5cde6004 100644
--- a/segmentation_models_pytorch/__init__.py
+++ b/segmentation_models_pytorch/__init__.py
@@ -26,7 +26,9 @@
 
 # Suppress the specific SyntaxWarning for `pretrainedmodels`
 warnings.filterwarnings("ignore", message="is with a literal", category=SyntaxWarning)
-warnings.filterwarnings("ignore", message=r'"is" with \'str\' literal.*', category=SyntaxWarning) # for python >= 3.12
+warnings.filterwarnings(
+    "ignore", message=r'"is" with \'str\' literal.*', category=SyntaxWarning
+)  # for python >= 3.12
 
 
 def create_model(
diff --git a/segmentation_models_pytorch/decoders/deeplabv3/model.py b/segmentation_models_pytorch/decoders/deeplabv3/model.py
index 030e42fd..73fc67e3 100644
--- a/segmentation_models_pytorch/decoders/deeplabv3/model.py
+++ b/segmentation_models_pytorch/decoders/deeplabv3/model.py
@@ -43,7 +43,7 @@ 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. 
+        kwargs: Arguments passed to the encoder class ``__init__()`` function. Applies only to ``timm`` models.
             Keys with ``None`` values are pruned before passing.
 
     Returns:
@@ -79,7 +79,7 @@ def __init__(
                     encoder_output_stride
                 )
             )
-        
+
         self.encoder = get_encoder(
             encoder_name,
             in_channels=in_channels,
@@ -91,7 +91,7 @@ def __init__(
 
         if upsampling is None:
             if encoder_depth <= 3:
-                scale_factor = 2 ** encoder_depth
+                scale_factor = 2**encoder_depth
             else:
                 scale_factor = encoder_output_stride
         else:
@@ -153,7 +153,7 @@ 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. 
+        kwargs: Arguments passed to the encoder class ``__init__()`` function. Applies only to ``timm`` models.
             Keys with ``None`` values are pruned before passing.
 
     Returns:

From c97d43a52ca0904dc033f180352156b26f5e1e32 Mon Sep 17 00:00:00 2001
From: Ryan <23580140+brianhou0208@users.noreply.github.com>
Date: Sat, 30 Nov 2024 03:38:57 +0800
Subject: [PATCH 3/7] Revert "fix ruff style"

This reverts commit 79d55685934e9d7496dea9a49d966b13165a0cf8.
---
 segmentation_models_pytorch/decoders/deeplabv3/model.py | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/segmentation_models_pytorch/decoders/deeplabv3/model.py b/segmentation_models_pytorch/decoders/deeplabv3/model.py
index 73fc67e3..030e42fd 100644
--- a/segmentation_models_pytorch/decoders/deeplabv3/model.py
+++ b/segmentation_models_pytorch/decoders/deeplabv3/model.py
@@ -43,7 +43,7 @@ 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.
+        kwargs: Arguments passed to the encoder class ``__init__()`` function. Applies only to ``timm`` models. 
             Keys with ``None`` values are pruned before passing.
 
     Returns:
@@ -79,7 +79,7 @@ def __init__(
                     encoder_output_stride
                 )
             )
-
+        
         self.encoder = get_encoder(
             encoder_name,
             in_channels=in_channels,
@@ -91,7 +91,7 @@ def __init__(
 
         if upsampling is None:
             if encoder_depth <= 3:
-                scale_factor = 2**encoder_depth
+                scale_factor = 2 ** encoder_depth
             else:
                 scale_factor = encoder_output_stride
         else:
@@ -153,7 +153,7 @@ 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.
+        kwargs: Arguments passed to the encoder class ``__init__()`` function. Applies only to ``timm`` models. 
             Keys with ``None`` values are pruned before passing.
 
     Returns:

From 14dc2a9bbf132e718dc05847b5bf2134283b62e8 Mon Sep 17 00:00:00 2001
From: Ryan <23580140+brianhou0208@users.noreply.github.com>
Date: Sat, 30 Nov 2024 02:11:36 +0800
Subject: [PATCH 4/7] fix encoder depth & output stride

---
 .../decoders/deeplabv3/decoder.py             | 41 +++++++------------
 .../decoders/deeplabv3/model.py               | 36 +++++++++++++---
 2 files changed, 44 insertions(+), 33 deletions(-)

diff --git a/segmentation_models_pytorch/decoders/deeplabv3/decoder.py b/segmentation_models_pytorch/decoders/deeplabv3/decoder.py
index e20acf3f..3fd73786 100644
--- a/segmentation_models_pytorch/decoders/deeplabv3/decoder.py
+++ b/segmentation_models_pytorch/decoders/deeplabv3/decoder.py
@@ -61,7 +61,6 @@ def __init__(
             nn.BatchNorm2d(out_channels),
             nn.ReLU(),
         )
-        self.out_channels = out_channels
 
     def forward(self, *features):
         return super().forward(features[-1])
@@ -79,17 +78,12 @@ def __init__(
         aspp_dropout: float,
     ):
         super().__init__()
-        if encoder_depth not in (3, 4, 5):
+        if encoder_depth < 3:
             raise ValueError(
-                "Encoder depth should be 3, 4 or 5, got {}.".format(encoder_depth)
+                "Encoder depth for DeepLabV3Plus decoder cannot be less than 3, got {}.".format(
+                    encoder_depth
+                )
             )
-        if output_stride not in (8, 16):
-            raise ValueError(
-                "Output stride should be 8 or 16, got {}.".format(output_stride)
-            )
-
-        self.out_channels = out_channels
-        self.output_stride = output_stride
 
         self.aspp = nn.Sequential(
             ASPP(
@@ -106,17 +100,10 @@ def __init__(
             nn.ReLU(),
         )
 
-        scale_factor = 2 if output_stride == 8 else 4
+        scale_factor = 4 if output_stride == 16 and encoder_depth > 3 else 2
         self.up = nn.UpsamplingBilinear2d(scale_factor=scale_factor)
 
-        if encoder_depth == 3 and output_stride == 8:
-            self.highres_input_index = -2
-        elif encoder_depth == 3 or encoder_depth == 4:
-            self.highres_input_index = -3
-        else:
-            self.highres_input_index = -4
-
-        highres_in_channels = encoder_channels[self.highres_input_index]
+        highres_in_channels = encoder_channels[2]
         highres_out_channels = 48  # proposed by authors of paper
         self.block1 = nn.Sequential(
             nn.Conv2d(
@@ -140,7 +127,7 @@ def __init__(
     def forward(self, *features):
         aspp_features = self.aspp(features[-1])
         aspp_features = self.up(aspp_features)
-        high_res_features = self.block1(features[self.highres_input_index])
+        high_res_features = self.block1(features[2])
         concat_features = torch.cat([aspp_features, high_res_features], dim=1)
         fused_features = self.block2(concat_features)
         return fused_features
@@ -240,13 +227,13 @@ def forward(self, x):
 class SeparableConv2d(nn.Sequential):
     def __init__(
         self,
-        in_channels,
-        out_channels,
-        kernel_size,
-        stride=1,
-        padding=0,
-        dilation=1,
-        bias=True,
+        in_channels: int,
+        out_channels: int,
+        kernel_size: int,
+        stride: int = 1,
+        padding: int = 0,
+        dilation: int = 1,
+        bias: bool = True,
     ):
         dephtwise_conv = nn.Conv2d(
             in_channels,
diff --git a/segmentation_models_pytorch/decoders/deeplabv3/model.py b/segmentation_models_pytorch/decoders/deeplabv3/model.py
index c0ef1238..6011019b 100644
--- a/segmentation_models_pytorch/decoders/deeplabv3/model.py
+++ b/segmentation_models_pytorch/decoders/deeplabv3/model.py
@@ -35,7 +35,7 @@ class DeepLabV3(SegmentationModel):
             Available options are **"sigmoid"**, **"softmax"**, **"logsoftmax"**, **"tanh"**, **"identity"**,
                 **callable** and **None**.
             Default is **None**
-        upsampling: Final upsampling factor (should have the same value as ``encoder_output_stride`` to preserve input-output spatial shape identity).
+        upsampling: Final upsampling factor. Default is **None** to preserve input-output spatial shape identity
         aux_params: Dictionary with parameters of the auxiliary output (classification head). Auxiliary output is build
             on top of encoder if **aux_params** is not **None** (default). Supported params:
                 - classes (int): A number of classes
@@ -43,7 +43,8 @@ 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.
+        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**
@@ -72,6 +73,13 @@ def __init__(
     ):
         super().__init__()
 
+        if encoder_output_stride not in [8, 16]:
+            raise ValueError(
+                "DeeplabV3 support output stride 8 or 16, got {}.".format(
+                    encoder_output_stride
+                )
+            )
+        
         self.encoder = get_encoder(
             encoder_name,
             in_channels=in_channels,
@@ -81,6 +89,14 @@ def __init__(
             **kwargs,
         )
 
+        if upsampling is None:
+            if encoder_depth <= 3:
+                scale_factor = 2 ** encoder_depth
+            else:
+                scale_factor = encoder_output_stride
+        else:
+            scale_factor = upsampling
+
         self.decoder = DeepLabV3Decoder(
             in_channels=self.encoder.out_channels[-1],
             out_channels=decoder_channels,
@@ -90,11 +106,11 @@ def __init__(
         )
 
         self.segmentation_head = SegmentationHead(
-            in_channels=self.decoder.out_channels,
+            in_channels=decoder_channels,
             out_channels=classes,
             activation=activation,
             kernel_size=1,
-            upsampling=encoder_output_stride if upsampling is None else upsampling,
+            upsampling=scale_factor,
         )
 
         if aux_params is not None:
@@ -138,7 +154,8 @@ 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.
+        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**
@@ -167,6 +184,13 @@ def __init__(
     ):
         super().__init__()
 
+        if encoder_output_stride not in [8, 16]:
+            raise ValueError(
+                "DeeplabV3Plus support output stride 8 or 16, got {}.".format(
+                    encoder_output_stride
+                )
+            )
+
         self.encoder = get_encoder(
             encoder_name,
             in_channels=in_channels,
@@ -187,7 +211,7 @@ def __init__(
         )
 
         self.segmentation_head = SegmentationHead(
-            in_channels=self.decoder.out_channels,
+            in_channels=decoder_channels,
             out_channels=classes,
             activation=activation,
             kernel_size=1,

From bafd4fbe2a4052de0acf66b112c1923a61c3c6f5 Mon Sep 17 00:00:00 2001
From: Ryan <23580140+brianhou0208@users.noreply.github.com>
Date: Sat, 30 Nov 2024 02:16:23 +0800
Subject: [PATCH 5/7] fix ruff style

---
 segmentation_models_pytorch/decoders/deeplabv3/model.py | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/segmentation_models_pytorch/decoders/deeplabv3/model.py b/segmentation_models_pytorch/decoders/deeplabv3/model.py
index 6011019b..9667415a 100644
--- a/segmentation_models_pytorch/decoders/deeplabv3/model.py
+++ b/segmentation_models_pytorch/decoders/deeplabv3/model.py
@@ -43,7 +43,7 @@ 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. 
+        kwargs: Arguments passed to the encoder class ``__init__()`` function. Applies only to ``timm`` models.
             Keys with ``None`` values are pruned before passing.
 
     Returns:
@@ -79,7 +79,7 @@ def __init__(
                     encoder_output_stride
                 )
             )
-        
+
         self.encoder = get_encoder(
             encoder_name,
             in_channels=in_channels,
@@ -91,7 +91,7 @@ def __init__(
 
         if upsampling is None:
             if encoder_depth <= 3:
-                scale_factor = 2 ** encoder_depth
+                scale_factor = 2**encoder_depth
             else:
                 scale_factor = encoder_output_stride
         else:
@@ -154,7 +154,7 @@ 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. 
+        kwargs: Arguments passed to the encoder class ``__init__()`` function. Applies only to ``timm`` models.
             Keys with ``None`` values are pruned before passing.
 
     Returns:

From f96dbda8c2091756faef7e424b008a1ee145e369 Mon Sep 17 00:00:00 2001
From: Ryan <23580140+brianhou0208@users.noreply.github.com>
Date: Sat, 30 Nov 2024 18:02:31 +0800
Subject: [PATCH 6/7] update deeplabv3+ doc

---
 segmentation_models_pytorch/decoders/deeplabv3/model.py | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/segmentation_models_pytorch/decoders/deeplabv3/model.py b/segmentation_models_pytorch/decoders/deeplabv3/model.py
index 9667415a..c5f366a2 100644
--- a/segmentation_models_pytorch/decoders/deeplabv3/model.py
+++ b/segmentation_models_pytorch/decoders/deeplabv3/model.py
@@ -145,9 +145,7 @@ class DeepLabV3Plus(SegmentationModel):
             Available options are **"sigmoid"**, **"softmax"**, **"logsoftmax"**, **"tanh"**, **"identity"**,
                 **callable** and **None**.
             Default is **None**
-        upsampling: Final upsampling factor. Default is 4 to preserve input-output spatial shape identity. In case
-            **encoder_depth** and **encoder_output_stride** are 3 and 16 resp., set **upsampling** to 2 to preserve.
-        aux_params: Dictionary with parameters of the auxiliary output (classification head). Auxiliary output is build
+        upsampling: Final upsampling factor. Default is 4 to preserve input-output spatial shape identity.
             on top of encoder if **aux_params** is not **None** (default). Supported params:
                 - classes (int): A number of classes
                 - pooling (str): One of "max", "avg". Default is "avg"

From 9fc8a4b35a0483e9215e1d8eb026f461d86fba28 Mon Sep 17 00:00:00 2001
From: Ryan <23580140+brianhou0208@users.noreply.github.com>
Date: Sat, 30 Nov 2024 23:26:36 +0800
Subject: [PATCH 7/7] restored aux_params

---
 segmentation_models_pytorch/decoders/deeplabv3/model.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/segmentation_models_pytorch/decoders/deeplabv3/model.py b/segmentation_models_pytorch/decoders/deeplabv3/model.py
index 53e1846e..830906cb 100644
--- a/segmentation_models_pytorch/decoders/deeplabv3/model.py
+++ b/segmentation_models_pytorch/decoders/deeplabv3/model.py
@@ -145,6 +145,7 @@ class DeepLabV3Plus(SegmentationModel):
                 **callable** and **None**.
             Default is **None**
         upsampling: Final upsampling factor. Default is 4 to preserve input-output spatial shape identity.
+        aux_params: Dictionary with parameters of the auxiliary output (classification head). Auxiliary output is build
             on top of encoder if **aux_params** is not **None** (default). Supported params:
                 - classes (int): A number of classes
                 - pooling (str): One of "max", "avg". Default is "avg"