Added 'cuda' functionality as well as streamlined

the dataloading process for segmentation tasks.

UniTrain/utils/classification.py

@@ -41,10 +41,8 @@ def get_data_loader(data_dir, batch_size, shuffle=True, transform = None, split=
 
     # Create a custom dataset
     dataset = ClassificationDataset(data_dir, transform=transform)
-    print(dataset.__len__())
 
     # Create a data loader
-    print(batch_size)
     data_loader = DataLoader(
         dataset,
         batch_size=batch_size,
@@ -86,7 +84,7 @@ def parse_folder(dataset_path):
         print("An error occurred:", str(e))
         return None
 
-def train_model(model, train_data_loader, test_data_loader, num_epochs, learning_rate=0.001, checkpoint_dir='checkpoints', logger=None, device='cpu'):
+def train_model(model, train_data_loader, test_data_loader, num_epochs, learning_rate=0.001, checkpoint_dir='checkpoints', logger=None, device=torch.device('cpu')):
     '''Train a PyTorch model for a classification task.
     Args:
     model (nn.Module): Torch model to train.
@@ -96,11 +94,17 @@ def train_model(model, train_data_loader, test_data_loader, num_epochs, learning
     learning_rate (float): Learning rate for the optimizer.
     checkpoint_dir (str): Directory to save model checkpoints.
     logger (Logger): Logger to log training details.
+    device (torch.device): Device to run training on (GPU or CPU).
 
     Returns:
     None
     '''
 
+    if logger:
+        logging.basicConfig(level=logging.INFO, format='%(asctime)s - Epoch %(epoch)d - Train Acc: %(train_acc).4f - Val Acc: %(val_acc).4f - %(message)s',
+                        datefmt='%Y-%m-%d %H:%M:%S', filename=logger, filemode='w')
+        logger = logging.getLogger(__name__)
+
     # Define loss function and optimizer
     criterion = nn.CrossEntropyLoss()
     optimizer = optim.Adam(model.parameters(), lr=learning_rate)
@@ -115,6 +119,9 @@ def train_model(model, train_data_loader, test_data_loader, num_epochs, learning
         for batch_idx, (inputs, labels) in enumerate(train_data_loader):
             optimizer.zero_grad()  # Zero the parameter gradients
 
+            inputs = inputs.to(device)
+            labels = labels.to(device)
+
             # Forward pass
             outputs = model(inputs)
             loss = criterion(outputs, labels)
@@ -138,7 +145,6 @@ def train_model(model, train_data_loader, test_data_loader, num_epochs, learning
             logger.info(f'Epoch {epoch + 1}, Validation Accuracy: {accuracy:.2f}%')
 
 
-
         if accuracy > best_accuracy:
             best_accuracy = accuracy
             checkpoint_path = os.path.join(checkpoint_dir, f'model_epoch_{epoch + 1}.pth')

UniTrain/utils/segmentation.py

@@ -7,9 +7,11 @@
 from torchvision import transforms
 from ..dataset.segmentation import SegmentationDataset
 import torchsummary
+import logging
+import glob
 
-def get_data_loader(image_paths:list, mask_paths:list, batch_size:int, shuffle:bool=True, transform=None) -> DataLoader:
-    """
+def get_data_loader(data_dir: str, batch_size:int, shuffle:bool=True, transform=None, split='train') -> DataLoader:
+    """,
     Create and return a data loader for a custom dataset.
 
     Args:
@@ -28,6 +30,9 @@ def get_data_loader(image_paths:list, mask_paths:list, batch_size:int, shuffle:b
             transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))  # Normalize with ImageNet stats
         ])
 
+    image_paths = glob.glob(os.path.join(data_dir, split, 'images', '*'))
+    mask_paths = glob.glob(os.path.join(data_dir, split, 'masks', '*'))
+
     # Create a custom dataset
     dataset = SegmentationDataset(image_paths=image_paths, mask_paths=mask_paths, transform=transform)
 
@@ -83,7 +88,7 @@ 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, iou=False, device='cpu'):
+def train_unet(model, train_data_loader, test_data_loader, num_epochs, learning_rate, checkpoint_dir, logger=None, iou=False, device=torch.device('cpu')) -> None:
     '''Train the model using the given train and test data loaders.
     
     Args: 
@@ -95,17 +100,17 @@ def train_unet(model, train_data_loader, test_data_loader, num_epochs, learning_
     checkpoint_dir (str): Directory to save model checkpoints.
     logger (Logger): Logger to log training information.
     iou (bool): Whether to calculate the IOU score.
+    device (torch.device): Device to run training on (GPU or CPU).
 
     Returns:
     None
     '''
-    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
+
+    if logger:
+        logging.basicConfig(level=logging.INFO, format='%(asctime)s - Epoch %(epoch)d - Train Acc: %(train_acc).4f - Val Acc: %(val_acc).4f - %(message)s',
+                        datefmt='%Y-%m-%d %H:%M:%S', filename=logger, filemode='w')
+        logger = logging.getLogger(__name__)
+
 
     criterion = nn.CrossEntropyLoss()
     optimizer = optim.Adam(model.parameters(), lr=learning_rate)
@@ -182,7 +187,6 @@ def iou_score(output, target):
     Returns:
     float: The average IoU score.
     '''
-
     smooth = 1e-6
     output = output.argmax(1)
     intersection = (output & target).float().sum((1, 2))

setup.py

@@ -2,7 +2,7 @@
 
 setup(
     name='UniTrain',
-    version='0.2.2',
+    version='0.2.3',
     author='Ishan Upadhyay',
     author_email='ishan.upadhyay.iitb@gmail.com',
     description='A generalized training framework for Deep Learning Tasks',