Merge Activation classes into one, added tanh (#315)

JulienMaille · Julien Maille · web-flow · commit 3ee899c35953 · 2020-12-17T13:38:32.000+03:00
* Merge Activation classes into one, added tanh

* update docstring

* minor typos

Co-authored-by: Julien Maille &lt;julien.maille@data-pixel.com&gt;
diff --git a/segmentation_models_pytorch/base/modules.py b/segmentation_models_pytorch/base/modules.py
@@ -89,14 +89,16 @@ def __init__(self, name, **params):
             self.activation = nn.Softmax(**params)
         elif name == 'logsoftmax':
             self.activation = nn.LogSoftmax(**params)
+        elif name == 'tanh':
+            self.activation = nn.Tanh()
         elif name == 'argmax':
             self.activation = ArgMax(**params)
         elif name == 'argmax2d':
             self.activation = ArgMax(dim=1, **params)
         elif callable(name):
             self.activation = name(**params)
         else:
-            raise ValueError('Activation should be callable/sigmoid/softmax/logsoftmax/None; got {}'.format(name))
+            raise ValueError('Activation should be callable/sigmoid/softmax/logsoftmax/tanh/None; got {}'.format(name))
 
     def forward(self, x):
         return self.activation(x)
diff --git a/segmentation_models_pytorch/deeplabv3/model.py b/segmentation_models_pytorch/deeplabv3/model.py
@@ -7,13 +7,13 @@
 
 
 class DeepLabV3(SegmentationModel):
-    """DeepLabV3_ implemetation from "Rethinking Atrous Convolution for Semantic Image Segmentation"
+    """DeepLabV3_ implementation from "Rethinking Atrous Convolution for Semantic Image Segmentation"
 
     Args:
         encoder_name: Name of the classification model that will be used as an encoder (a.k.a backbone)
             to extract features of different spatial resolution
         encoder_depth: A number of stages used in encoder in range [3, 5]. Each stage generate features 
-            two times smaller in spatial dimentions than previous one (e.g. for depth 0 we will have features 
+            two times smaller in spatial dimensions than previous one (e.g. for depth 0 we will have features
             with shapes [(N, C, H, W),], for depth 1 - [(N, C, H, W), (N, C, H // 2, W // 2)] and so on).
             Default is 5
         encoder_weights: One of **None** (random initialization), **"imagenet"** (pre-training on ImageNet) and 
@@ -22,7 +22,7 @@ class DeepLabV3(SegmentationModel):
         in_channels: A number of input channels for the model, default is 3 (RGB images)
         classes: A number of classes for output mask (or you can think as a number of channels of output mask)
         activation: An activation function to apply after the final convolution layer.
-            Avaliable options are **"sigmoid"**, **"softmax"**, **"logsoftmax"**, **"identity"**, **callable** and **None**.
+            Available options are **"sigmoid"**, **"softmax"**, **"logsoftmax"**, **"tanh"**, **"identity"**, **callable** and **None**.
             Default is **None**
         upsampling: Final upsampling factor. Default is 8 to preserve input-output spatial shape identity
         aux_params: Dictionary with parameters of the auxiliary output (classification head). Auxiliary output is build 
@@ -86,14 +86,14 @@ def __init__(
 
 
 class DeepLabV3Plus(SegmentationModel):
-    """DeepLabV3+ implemetation from "Encoder-Decoder with Atrous Separable
+    """DeepLabV3+ implementation from "Encoder-Decoder with Atrous Separable
     Convolution for Semantic Image Segmentation"
     
     Args:
         encoder_name: Name of the classification model that will be used as an encoder (a.k.a backbone)
             to extract features of different spatial resolution
         encoder_depth: A number of stages used in encoder in range [3, 5]. Each stage generate features 
-            two times smaller in spatial dimentions than previous one (e.g. for depth 0 we will have features 
+            two times smaller in spatial dimensions than previous one (e.g. for depth 0 we will have features
             with shapes [(N, C, H, W),], for depth 1 - [(N, C, H, W), (N, C, H // 2, W // 2)] and so on).
             Default is 5
         encoder_weights: One of **None** (random initialization), **"imagenet"** (pre-training on ImageNet) and 
@@ -104,7 +104,7 @@ class DeepLabV3Plus(SegmentationModel):
         in_channels: A number of input channels for the model, default is 3 (RGB images)
         classes: A number of classes for output mask (or you can think as a number of channels of output mask)
         activation: An activation function to apply after the final convolution layer.
-            Avaliable options are **"sigmoid"**, **"softmax"**, **"logsoftmax"**, **"identity"**, **callable** and **None**.
+            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
         aux_params: Dictionary with parameters of the auxiliary output (classification head). Auxiliary output is build 
diff --git a/segmentation_models_pytorch/encoders/__init__.py b/segmentation_models_pytorch/encoders/__init__.py
@@ -51,7 +51,7 @@ def get_encoder(name, in_channels=3, depth=5, weights=None):
         try:
             settings = encoders[name]["pretrained_settings"][weights]
         except KeyError:
-            raise KeyError("Wrong pretrained weights `{}` for encoder `{}`. Avaliable options are: {}".format(
+            raise KeyError("Wrong pretrained weights `{}` for encoder `{}`. Available options are: {}".format(
                 weights, name, list(encoders[name]["pretrained_settings"].keys()),
             ))
         encoder.load_state_dict(model_zoo.load_url(settings["url"]))
@@ -69,7 +69,7 @@ def get_preprocessing_params(encoder_name, pretrained="imagenet"):
     settings = encoders[encoder_name]["pretrained_settings"]
 
     if pretrained not in settings.keys():
-        raise ValueError("Avaliable pretrained options {}".format(settings.keys()))
+        raise ValueError("Available pretrained options {}".format(settings.keys()))
 
     formatted_settings = {}
     formatted_settings["input_space"] = settings[pretrained].get("input_space")
diff --git a/segmentation_models_pytorch/encoders/_base.py b/segmentation_models_pytorch/encoders/_base.py
@@ -18,7 +18,7 @@ def out_channels(self):
         return self._out_channels[: self._depth + 1]
 
     def set_in_channels(self, in_channels):
-        """Change first convolution chennels"""
+        """Change first convolution channels"""
         if in_channels == 3:
             return
 
diff --git a/segmentation_models_pytorch/fpn/model.py b/segmentation_models_pytorch/fpn/model.py
@@ -11,19 +11,19 @@ class FPN(SegmentationModel):
         encoder_name: Name of the classification model that will be used as an encoder (a.k.a backbone)
             to extract features of different spatial resolution
         encoder_depth: A number of stages used in encoder in range [3, 5]. Each stage generate features 
-            two times smaller in spatial dimentions than previous one (e.g. for depth 0 we will have features 
+            two times smaller in spatial dimensions than previous one (e.g. for depth 0 we will have features
             with shapes [(N, C, H, W),], for depth 1 - [(N, C, H, W), (N, C, H // 2, W // 2)] and so on).
             Default is 5
         encoder_weights: One of **None** (random initialization), **"imagenet"** (pre-training on ImageNet) and 
             other pretrained weights (see table with available weights for each encoder_name)
         decoder_pyramid_channels: A number of convolution filters in Feature Pyramid of FPN_
         decoder_segmentation_channels: A number of convolution filters in segmentation blocks of FPN_
-        decoder_merge_policy: Determines how to merge pyramid features inside FPN. Avaliable options are **add** and **cat**
+        decoder_merge_policy: Determines how to merge pyramid features inside FPN. Available options are **add** and **cat**
         decoder_dropout: Spatial dropout rate in range (0, 1) for feature pyramid in FPN_
         in_channels: A number of input channels for the model, default is 3 (RGB images)
         classes: A number of classes for output mask (or you can think as a number of channels of output mask)
         activation: An activation function to apply after the final convolution layer.
-            Avaliable options are **"sigmoid"**, **"softmax"**, **"logsoftmax"**, **"identity"**, **callable** and **None**.
+            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
         aux_params: Dictionary with parameters of the auxiliary output (classification head). Auxiliary output is build 
diff --git a/segmentation_models_pytorch/linknet/model.py b/segmentation_models_pytorch/linknet/model.py
@@ -17,18 +17,18 @@ class Linknet(SegmentationModel):
         encoder_name: Name of the classification model that will be used as an encoder (a.k.a backbone)
             to extract features of different spatial resolution
         encoder_depth: A number of stages used in encoder in range [3, 5]. Each stage generate features 
-            two times smaller in spatial dimentions than previous one (e.g. for depth 0 we will have features 
+            two times smaller in spatial dimensions than previous one (e.g. for depth 0 we will have features
             with shapes [(N, C, H, W),], for depth 1 - [(N, C, H, W), (N, C, H // 2, W // 2)] and so on).
             Default is 5
         encoder_weights: One of **None** (random initialization), **"imagenet"** (pre-training on ImageNet) and 
             other pretrained weights (see table with available weights for each encoder_name)
         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.
-            Avaliable options are **True, False, "inplace"**
+            Available options are **True, False, "inplace"**
         in_channels: A number of input channels for the model, default is 3 (RGB images)
         classes: A number of classes for output mask (or you can think as a number of channels of output mask)
         activation: An activation function to apply after the final convolution layer.
-            Avaliable options are **"sigmoid"**, **"softmax"**, **"logsoftmax"**, **"identity"**, **callable** and **None**.
+            Available options are **"sigmoid"**, **"softmax"**, **"logsoftmax"**, **"tanh"**, **"identity"**, **callable** and **None**.
             Default is **None**
         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:
diff --git a/segmentation_models_pytorch/manet/model.py b/segmentation_models_pytorch/manet/model.py
@@ -16,22 +16,22 @@ class MAnet(SegmentationModel):
         encoder_name: Name of the classification model that will be used as an encoder (a.k.a backbone)
             to extract features of different spatial resolution
         encoder_depth: A number of stages used in encoder in range [3, 5]. Each stage generate features 
-            two times smaller in spatial dimentions than previous one (e.g. for depth 0 we will have features 
+            two times smaller in spatial dimensions than previous one (e.g. for depth 0 we will have features
             with shapes [(N, C, H, W),], for depth 1 - [(N, C, H, W), (N, C, H // 2, W // 2)] and so on).
             Default is 5
         encoder_weights: One of **None** (random initialization), **"imagenet"** (pre-training on ImageNet) and 
             other pretrained weights (see table with available weights for each encoder_name)
         decoder_channels: List of integers which specify **in_channels** parameter for convolutions used in decoder.
-            Lenght of the list should be the same as **encoder_depth**
+            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.
-            Avaliable options are **True, False, "inplace"**
+            Available options are **True, False, "inplace"**
         decoder_pab_channels: A number of channels for PAB module in decoder. 
             Default is 64.
         in_channels: A number of input channels for the model, default is 3 (RGB images)
         classes: A number of classes for output mask (or you can think as a number of channels of output mask)
         activation: An activation function to apply after the final convolution layer.
-            Avaliable options are **"sigmoid"**, **"softmax"**, **"logsoftmax"**, **"identity"**, **callable** and **None**.
+            Available options are **"sigmoid"**, **"softmax"**, **"logsoftmax"**, **"tanh"**, **"identity"**, **callable** and **None**.
             Default is **None**
         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:
diff --git a/segmentation_models_pytorch/pan/model.py b/segmentation_models_pytorch/pan/model.py
@@ -23,7 +23,7 @@ class PAN(SegmentationModel):
         in_channels: A number of input channels for the model, default is 3 (RGB images)
         classes: A number of classes for output mask (or you can think as a number of channels of output mask)
         activation: An activation function to apply after the final convolution layer.
-            Avaliable options are **"sigmoid"**, **"softmax"**, **"logsoftmax"**, **"identity"**, **callable** and **None**.
+            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
         aux_params: Dictionary with parameters of the auxiliary output (classification head). Auxiliary output is build 
diff --git a/segmentation_models_pytorch/pspnet/model.py b/segmentation_models_pytorch/pspnet/model.py
@@ -17,20 +17,20 @@ class PSPNet(SegmentationModel):
         encoder_name: Name of the classification model that will be used as an encoder (a.k.a backbone)
             to extract features of different spatial resolution
         encoder_depth: A number of stages used in encoder in range [3, 5]. Each stage generate features 
-            two times smaller in spatial dimentions than previous one (e.g. for depth 0 we will have features 
+            two times smaller in spatial dimensions than previous one (e.g. for depth 0 we will have features
             with shapes [(N, C, H, W),], for depth 1 - [(N, C, H, W), (N, C, H // 2, W // 2)] and so on).
             Default is 5
         encoder_weights: One of **None** (random initialization), **"imagenet"** (pre-training on ImageNet) and 
             other pretrained weights (see table with available weights for each encoder_name)
-        psp_out_channels: A number of filters in Saptial Pyramid
+        psp_out_channels: A number of filters in Spatial Pyramid
         psp_use_batchnorm: If **True**, BatchNorm2d layer between Conv2D and Activation layers
             is used. If **"inplace"** InplaceABN will be used, allows to decrease memory consumption.
-            Avaliable options are **True, False, "inplace"**
+            Available options are **True, False, "inplace"**
         psp_dropout: Spatial dropout rate in [0, 1) used in Spatial Pyramid
         in_channels: A number of input channels for the model, default is 3 (RGB images)
         classes: A number of classes for output mask (or you can think as a number of channels of output mask)
         activation: An activation function to apply after the final convolution layer.
-            Avaliable options are **"sigmoid"**, **"softmax"**, **"logsoftmax"**, **"identity"**, **callable** and **None**.
+            Available options are **"sigmoid"**, **"softmax"**, **"logsoftmax"**, **"tanh"**, **"identity"**, **callable** and **None**.
             Default is **None**
         upsampling: Final upsampling factor. Default is 8 to preserve input-output spatial shape identity
         aux_params: Dictionary with parameters of the auxiliary output (classification head). Auxiliary output is build 
diff --git a/segmentation_models_pytorch/unet/model.py b/segmentation_models_pytorch/unet/model.py
@@ -15,22 +15,22 @@ class Unet(SegmentationModel):
         encoder_name: Name of the classification model that will be used as an encoder (a.k.a backbone)
             to extract features of different spatial resolution
         encoder_depth: A number of stages used in encoder in range [3, 5]. Each stage generate features 
-            two times smaller in spatial dimentions than previous one (e.g. for depth 0 we will have features 
+            two times smaller in spatial dimensions than previous one (e.g. for depth 0 we will have features
             with shapes [(N, C, H, W),], for depth 1 - [(N, C, H, W), (N, C, H // 2, W // 2)] and so on).
             Default is 5
         encoder_weights: One of **None** (random initialization), **"imagenet"** (pre-training on ImageNet) and 
             other pretrained weights (see table with available weights for each encoder_name)
         decoder_channels: List of integers which specify **in_channels** parameter for convolutions used in decoder.
-            Lenght of the list should be the same as **encoder_depth**
+            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.
-            Avaliable options are **True, False, "inplace"**
-        decoder_attention_type: Attention module used in decoder of the model. Avaliable options are **None** and **scse**.
+            Available options are **True, False, "inplace"**
+        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
         in_channels: A number of input channels for the model, default is 3 (RGB images)
         classes: A number of classes for output mask (or you can think as a number of channels of output mask)
         activation: An activation function to apply after the final convolution layer.
-            Avaliable options are **"sigmoid"**, **"softmax"**, **"logsoftmax"**, **"identity"**, **callable** and **None**.
+            Available options are **"sigmoid"**, **"softmax"**, **"logsoftmax"**, **"tanh"**, **"identity"**, **callable** and **None**.
             Default is **None**
         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:
diff --git a/segmentation_models_pytorch/unetplusplus/decoder.py b/segmentation_models_pytorch/unetplusplus/decoder.py
@@ -119,7 +119,7 @@ def forward(self, *features):
 
         features = features[1:]    # remove first skip with same spatial resolution
         features = features[::-1]  # reverse channels to start from head of encoder
-        # start bulding dense connections
+        # start building dense connections
         dense_x = {}
         for layer_idx in range(len(self.in_channels)-1):
             for depth_idx in range(self.depth-layer_idx):
diff --git a/segmentation_models_pytorch/unetplusplus/model.py b/segmentation_models_pytorch/unetplusplus/model.py
diff --git a/segmentation_models_pytorch/utils/base.py b/segmentation_models_pytorch/utils/base.py
diff --git a/segmentation_models_pytorch/utils/losses.py b/segmentation_models_pytorch/utils/losses.py
diff --git a/segmentation_models_pytorch/utils/meter.py b/segmentation_models_pytorch/utils/meter.py
diff --git a/segmentation_models_pytorch/utils/metrics.py b/segmentation_models_pytorch/utils/metrics.py
