new_feature: Added functionality 1. IOU Score 2.Choosing devices

Segmentation/train.py

@@ -23,6 +23,8 @@ def main():
     parser.add_argument('--logging_directory', type=str, default='logs', help='Directory for logging')
     parser.add_argument('--checkpoint_directory', type=str, default='checkpoints', help='Directory for saving checkpoints')
     parser.add_argument('--classes', type=int, default='2', help='No. of classes you want to segment your model into.')
+    parser.add_argument('--iou', type=bool, default=False, help='Enable or disable IoU')
+    parser.add_argument('--device', type=str, default='cpu', help='Device to train on')
     args = parser.parse_args()
 
     # Create the logging directory
@@ -94,7 +96,9 @@ def main():
         num_epochs=args.epochs,
         learning_rate=args.learning_rate,
         checkpoint_dir=args.checkpoint_directory,
-        logger=logging
+        logger=logging,
+        iou=args.iou,
+        device=args.device
     )
 
 

Segmentation/utils.py

@@ -75,7 +75,15 @@ def parse_folder(dataset_path):
         return False
 
 
-def train_unet(model, train_data_loader, test_data_loader, num_epochs, learning_rate, checkpoint_dir, logger=None):
+def train_unet(model, train_data_loader, test_data_loader, num_epochs, learning_rate, checkpoint_dir, logger=None, iou=False, device='cpu'):
+    if device == 'cpu':
+        device = torch.device('cpu')
+    elif device == 'cuda':
+        device = torch.device('cuda')
+    else:
+        print(f"{device} is not a valid device.")
+        return None
+
     criterion = nn.CrossEntropyLoss()
     optimizer = optim.Adam(model.parameters(), lr=learning_rate)
 
@@ -84,36 +92,49 @@ def train_unet(model, train_data_loader, test_data_loader, num_epochs, learning_
     for epoch in range(num_epochs):
         model.train()
         train_loss = 0.0
+        iou_score_mean = 0.0
 
         for inputs, targets in tqdm(train_data_loader, desc=f'Epoch {epoch + 1}/{num_epochs}', leave=False):
             optimizer.zero_grad()
             outputs = model(inputs)
             targets = targets.squeeze(1)
+            outputs.to(device)
+            targets.to(device)
             loss = criterion(outputs, targets)
             loss.backward()
             optimizer.step()
             train_loss += loss.item()
+            iou_score_mean += iou_score(outputs, targets)
 
+        iou_score_mean = iou_score_mean / len(train_data_loader)
         average_train_loss = train_loss / len(train_data_loader)
 
-        if logger:
-            logger.info(f'Epoch {epoch + 1}/{num_epochs}, Train Loss: {average_train_loss:.4f}')
+        if logger and iou:
+            logger.info(f'Epoch {epoch + 1}/{num_epochs}, Train Loss: {average_train_loss:.4f}, IOU Score: {iou_score_mean:.4f}')
+        else:
+            logger.info(f"Epoch {epoch + 1}/{num_epochs}, Train Loss: {average_train_loss:.4f}")
 
         # Validation
         model.eval()
         val_loss = 0.0
-
+        iou_score_mean = 0.0
         with torch.no_grad():
             for inputs, targets in tqdm(test_data_loader, desc=f'Validation', leave=False):
                 outputs = model(inputs)
                 targets = targets.squeeze(1)
+                outputs.to(device)
+                targets.to(device)
                 loss = criterion(outputs, targets)
                 val_loss += loss.item()
+                iou_score_mean += iou_score(outputs, targets)
 
+        iou_score_mean = iou_score_mean / len(test_data_loader)
         average_val_loss = val_loss / len(test_data_loader)
 
-        if logger:
-            logger.info(f'Epoch {epoch + 1}/{num_epochs}, Validation Loss: {average_val_loss:.4f}')
+        if logger and iou:
+            logger.info(f'Epoch {epoch + 1}/{num_epochs}, Validation Loss: {average_val_loss:.4f}. IOU Score: {iou_score_mean:.4f}')
+        else:
+            logger.info(f"Epoch {epoch + 1}/{num_epochs}, Validation Loss: {average_val_loss:.4f}")
 
         # Save model checkpoint if validation loss improves
         if average_val_loss < best_loss:
@@ -126,4 +147,12 @@ def train_unet(model, train_data_loader, test_data_loader, num_epochs, learning_
     print('Finished Training')
 
 def generate_model_summary(model, input_size):
-    torchsummary.summary(model, input_size=input_size)
+    torchsummary.summary(model, input_size=input_size)
+
+def iou_score(output, target):
+    smooth = 1e-6
+    output = output.argmax(1)
+    intersection = (output & target).float().sum((1, 2))
+    union = (output | target).float().sum((1, 2))
+    iou = (intersection + smooth) / (union + smooth)
+    return iou.mean()