Merge pull request #81 from SFI-Visual-Intelligence/solveig-develope

added nr_channels to dataloader and updated modelto work with any input size

Author: Seilmast

CollaborativeCoding/dataloaders/uspsh5_7_9.py

@@ -32,7 +32,7 @@ class USPSH5_Digit_7_9_Dataset(Dataset):
         A transform function to apply to the images.
     """
 
-    def __init__(self, data_path, train=False, transform=None):
+    def __init__(self, data_path, sample_ids, train=False, transform=None,  nr_channels=1):
         super().__init__()
         """
         Initializes the USPS dataset by loading images and labels from the given `.h5` file.
@@ -51,6 +51,8 @@ def __init__(self, data_path, train=False, transform=None):
         self.transform = transform
         self.mode = "train" if train else "test"
         self.h5_path = data_path / self.filename
+        self.sample_ids = sample_ids
+        self.nr_channels = nr_channels
 
         # Load the dataset from the HDF5 file
         with h5py.File(self.filepath, "r") as hf:
@@ -107,10 +109,10 @@ def main():
             transforms.Normalize((0.5,), (0.5,)),  # Normalize to [-1, 1]
         ]
     )
-
+    indices = np.array([7, 8, 9])
     # Load the dataset
     dataset = USPSH5_Digit_7_9_Dataset(
-        data_path="C:/Users/Solveig/OneDrive/Dokumente/UiT PhD/Courses/Git",
+        data_path="C:/Users/Solveig/OneDrive/Dokumente/UiT PhD/Courses/Git", sample_ids=indices,
         train=False,
         transform=transform,
     )

CollaborativeCoding/models/solveig_model.py

@@ -2,6 +2,37 @@
 import torch.nn as nn
 
 
+def find_fc_input_shape(image_shape, model):
+    """
+      Find the shape of the input to the fully connected layer after passing through the convolutional layers.
+
+      Code inspired by @Seilmast (https://github.com/SFI-Visual-Intelligence/Collaborative-Coding-Exam/issues/67#issuecomment-2651212254)
+
+      Args
+      ----
+      image_shape : tuple(int, int, int)
+          Shape of the input image (C, H, W), where C is the number of channels,
+          H is the height, and W is the width of the image.
+      model : nn.Module
+          The CNN model containing the convolutional layers, whose output size is used to
+          determine the number of input features for the fully connected layer.
+
+      Returns
+      -------
+      int
+          The number of elements in the input to the fully connected layer.
+      """
+
+    dummy_img = torch.randn(1, *image_shape)
+    with torch.no_grad():
+        x = model.conv_block1(dummy_img)
+        x = model.conv_block2(x)
+        x = model.conv_block3(x)
+        x = torch.flatten(x, 1)
+
+    return x.size(1)
+
+
 class SolveigModel(nn.Module):
     """
     A Convolutional Neural Network model for classification.
@@ -49,9 +80,19 @@ def __init__(self, image_shape, num_classes):
             nn.ReLU(),
         )
 
-        self.fc1 = nn.Linear(100 * 8 * 8, num_classes)
+        fc_input_size = find_fc_input_shape(image_shape, self)
+
+        self.fc1 = nn.Linear(fc_input_size, num_classes)
 
     def forward(self, x):
+        """
+        Defines the forward pass.
+        Args:
+            x (torch.Tensor): A four-dimensional tensor with shape
+                        (Batch Size, Channels, Image Height, Image Width).
+        Returns:
+        torch.Tensor: The output tensor containing class logits for each input sample.
+        """
         x = self.conv_block1(x)
         x = self.conv_block2(x)
         x = self.conv_block3(x)
@@ -63,7 +104,7 @@ def forward(self, x):
 
 
 if __name__ == "__main__":
-    x = torch.randn(1, 3, 16, 16)
+    x = torch.randn(1, 3, 28, 28)
 
     model = SolveigModel(x.shape[1:], 3)