Fix encoder depth & output stride on DeeplabV3 & DeeplabV3+

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,

segmentation_models_pytorch/decoders/deeplabv3/model.py

@@ -35,15 +35,16 @@ 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
                 - pooling (str): One of "max", "avg". Default is "avg"
                 - 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 +73,12 @@ 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 +88,14 @@ 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 +105,11 @@ 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:
@@ -129,16 +144,16 @@ 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.
+        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"
                 - 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 +182,13 @@ 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 +209,7 @@ def __init__(
         )
 
         self.segmentation_head = SegmentationHead(
-            in_channels=self.decoder.out_channels,
+            in_channels=decoder_channels,
             out_channels=classes,
             activation=activation,
             kernel_size=1,