Add resnet

src/__init__.py

@@ -9,7 +9,7 @@
 
 
 # Register models
-registry.Model(cell_resnet)
+registry.Model(ResNet)
 registry.Model(cell_senet)
 registry.Model(cell_densenet)
 registry.Model(EfficientNet)

src/models/__init__.py

@@ -1,4 +1,4 @@
-from .resnet import cell_resnet
+from .resnet import ResNet
 from .senet import cell_senet
 from .densenet import cell_densenet
 from .efficientnet import EfficientNet

src/models/resnet.py

@@ -1,30 +1,39 @@
 import torch.nn as nn
 import pretrainedmodels
 from cnn_finetune import make_model
+import timm
+from .utils import *
 
 
-def cell_resnet(model_name, num_classes=1108, n_channels=6):
-    model = make_model(
-        model_name=model_name,
-        num_classes=num_classes,
-        pretrained=True
-    )
-    conv1 = model._features[0]
-    model._features[0] = nn.Conv2d(in_channels=n_channels,
-                            out_channels=conv1.out_channels,
-                            kernel_size=conv1.kernel_size,
-                            stride=conv1.stride,
-                            padding=conv1.padding,
-                            bias=conv1.bias)
-
-    # copy pretrained weights
-    model._features[0].weight.data[:,:3,:,:] = conv1.weight.data
-    model._features[0].weight.data[:,3:n_channels,:,:] = conv1.weight.data[:,:int(n_channels-3),:,:]
-    return model
-
-
-if __name__ == '__main__':
-    import torch
-    model = cell_resnet(model_name='resnet18')
-    x = torch.randn((1, 6, 320, 320))
-    y = model(x)
+class ResNet(nn.Module):
+    def __init__(self,  model_name="resnet50",
+                        num_classes=1108,
+                        n_channels=6):
+        super(ResNet, self).__init__()
+
+        self.model = timm.create_model(model_name, pretrained=True, num_classes=num_classes)
+        conv1 = self.model.conv1
+        self.model.conv1 = nn.Conv2d(in_channels=n_channels,
+                                             out_channels=conv1.out_channels,
+                                             kernel_size=conv1.kernel_size,
+                                             stride=conv1.stride,
+                                             padding=conv1.padding,
+                                             bias=conv1.bias)
+
+        # copy pretrained weights
+        self.model.conv1.weight.data[:, :3, :, :] = conv1.weight.data
+        self.model.conv1.weight.data[:, 3:n_channels, :, :] = conv1.weight.data[:, :int(n_channels - 3), :, :]
+
+    def forward(self, x):
+        return self.model(x)
+
+    def freeze(self):
+        for param in self.model.parameters():
+            param.requires_grad = False
+
+        for param in self.model.fc.parameters():
+            param.requires_grad = True
+
+    def unfreeze(self):
+        for param in self.model.parameters():
+            param.requires_grad = True