Merge pull request #23 from mdeleeuw1/master

Multiloss Pytorch implementation

hooknet/models/torchmodel.py

@@ -2,7 +2,6 @@
 import torch.nn as nn
 from torchvision.transforms.functional import center_crop
 
-
 class HookNet(nn.Module):
     def __init__(
         self,
@@ -45,13 +44,18 @@ def __init__(
             hook_channels=mid_channels,
             hook_to_index=3,
         )
-        self.last_conv = nn.Conv2d(self.high_mag_branch.decoder._out_channels[0], n_classes, 1)
+
+        self.high_last_conv = nn.Conv2d(self.high_mag_branch.decoder._out_channels[0], n_classes, 1)
+        self.mid_last_conv = nn.Conv2d(self.mid_mag_branch.decoder._out_channels[0], n_classes, 1)
+        self.low_last_conv = nn.Conv2d(self.low_mag_branch.decoder._out_channels[0], n_classes, 1)
 
     def forward(self, high_input, mid_input, low_input):
-        low_out = self.low_mag_branch(low_input)
-        mid_out = self.mid_mag_branch(mid_input, low_out)
-        high_out = self.high_mag_branch(high_input, mid_out)
-        return {'out': self.last_conv(high_out)}
+        low_out, low_hook_out = self.low_mag_branch(low_input)
+        mid_out, mid_hook_out = self.mid_mag_branch(mid_input, low_hook_out)
+        high_out, high_hook_out = self.high_mag_branch(high_input, mid_hook_out)
+        return {'high_out': self.high_last_conv(high_out),
+                'mid_out': self.mid_last_conv(mid_out),
+                'low_out': self.low_last_conv(low_out)}
 
 
 class Branch(nn.Module):
@@ -91,8 +95,8 @@ def __init__(
     def forward(self, x, hook_in=None):
         out, residuals = self.encoder(x)
         out = self.mid_conv_block(out)
-        out = self.decoder(out, residuals, hook_in)
-        return out
+        out, hook_out = self.decoder(out, residuals, hook_in)
+        return out, hook_out
 
 
 class Encoder(nn.Module):
@@ -156,18 +160,22 @@ def __init__(
 
     def forward(self, x, residuals, hook_in=None):
         out = x
+        hook_true = False
         for d in reversed(range(self._depth)):
             if hook_in is not None and d == self._hook_to_index:
                 out = concatenator(out, hook_in)
 
             out = self._decode_path[f"upsample{d}"](out)
             out = concatenator(out, residuals[d])
             out = self._decode_path[f"convblock{d}"](out)
-
+            
             if self._hook_from_index is not None and d == self._hook_from_index:
-                return out
+                hook_out = out
+                hook_true = True
 
-        return out
+        if hook_true == False:
+            hook_out = out
+        return out, hook_out
 
 
 class ConvBlock(nn.Module):
@@ -211,4 +219,4 @@ def forward(self, x):
 def concatenator(x, x2):
     x2_cropped = center_crop(x2, x.shape[-1])
     conc = torch.cat([x, x2_cropped], dim=1)
-    return conc
+    return conc