Making ASPP-Layer in DeepLab more generic (#2174)

At the moment in the ASPP-Layer the number of output channels are predefined as a constant,
which is good for DeepLab but not necessairly in other projects, where another out-channel Nr. is required.

Also the number of "atrous rates" is fixed to three, which also could be sometimes more or less depending on the notwork-arch.
Again these fixed values may make sense in DeepLab-Model but not necessarily in other type of models.

This pull-req. contains the needed changes to make ASPP-Layer generic.

Author: ArashJavan

torchvision/models/segmentation/deeplabv3.py

@@ -63,19 +63,18 @@ def forward(self, x):
 
 
 class ASPP(nn.Module):
-    def __init__(self, in_channels, atrous_rates):
+    def __init__(self, in_channels, atrous_rates, out_channels=256):
         super(ASPP, self).__init__()
-        out_channels = 256
         modules = []
         modules.append(nn.Sequential(
             nn.Conv2d(in_channels, out_channels, 1, bias=False),
             nn.BatchNorm2d(out_channels),
             nn.ReLU()))
 
-        rate1, rate2, rate3 = tuple(atrous_rates)
-        modules.append(ASPPConv(in_channels, out_channels, rate1))
-        modules.append(ASPPConv(in_channels, out_channels, rate2))
-        modules.append(ASPPConv(in_channels, out_channels, rate3))
+        rates = tuple(atrous_rates)
+        for rate in rates:
+            modules.append(ASPPConv(in_channels, out_channels, rate))
+
         modules.append(ASPPPooling(in_channels, out_channels))
 
         self.convs = nn.ModuleList(modules)