Improve structure and readability for FCNHead (open-mmlab#2142)

Author: edwardyehuang

mmseg/models/decode_heads/fcn_head.py

@@ -37,28 +37,21 @@ def __init__(self,
 
         conv_padding = (kernel_size // 2) * dilation
         convs = []
-        convs.append(
-            ConvModule(
-                self.in_channels,
-                self.channels,
-                kernel_size=kernel_size,
-                padding=conv_padding,
-                dilation=dilation,
-                conv_cfg=self.conv_cfg,
-                norm_cfg=self.norm_cfg,
-                act_cfg=self.act_cfg))
-        for i in range(num_convs - 1):
+        for i in range(num_convs):
+            _in_channels = self.in_channels if i == 0 else self.channels
             convs.append(
                 ConvModule(
-                    self.channels,
+                    _in_channels,
                     self.channels,
                     kernel_size=kernel_size,
                     padding=conv_padding,
                     dilation=dilation,
                     conv_cfg=self.conv_cfg,
                     norm_cfg=self.norm_cfg,
-                    act_cfg=self.act_cfg))
-        if num_convs == 0:
+                    act_cfg=self.act_cfg
+                ))
+            
+        if len(convs) == 0:
             self.convs = nn.Identity()
         else:
             self.convs = nn.Sequential(*convs)