More U-Net Like UOLO (ultralytics#19)

models/common.py

@@ -852,36 +852,32 @@ class Seg(nn.Module):
     def __init__(self, in_channels):
         super().__init__()
         print('SEG in channels: ', in_channels)
-        self.cv1 = Conv(in_channels, 8, k=3)
+        self.cv1 = Conv(in_channels, 96, k=3)
         self.upsample = nn.Upsample(scale_factor=2, mode='nearest')
-        self.cv2 = Conv(8, 16, k=3)
-        self.cv3 = Conv(16, 8, k=3)
-        self.cv4 = Conv(8, 1, act=False)
+        self.cv2 = Conv(192, 48, k=3)
+        self.cv3 = Conv(48, 16, k=3)
+        self.cv4 = Conv(16, 1, act=False)
         self.relu = nn.ReLU()
 
         # self.dropout_weak = nn.Dropout(0.25)
         self.dropout_normal = nn.Dropout(0.5)
         # self.sigmoid = nn.Sigmoid()
 
-    def forward(self, x, new_x):
-        print('new x:', new_x.shape)
+    def forward(self, x, skipped_input):
         # print('----entry shape', x.shape, '---\n')
         x = self.cv1(x)
         x = self.upsample(x)
-        print('post unsample 1:', x.shape)
-        # x = self.relu(x)
+        # Here you could use 2/1  96 - channels
+        x = torch.cat((x, skipped_input), 1)
+
         # print('----upsample shape', x.shape, '---\n')
         x = self.cv2(x)
         x = self.upsample(x)
-        print('post unsample 2:', x.shape)
 
-        # x = self.relu(x)
         # x = self.dropout_normal(x)
         x = self.cv3(x)
         x = self.upsample(x)
-        print('post unsample 3:', x.shape)
         # print('----out shape', x.shape, '---\n')
-        # x = self.sigmoid(x)
 
         x = self.cv4(x)
         return x