diff --git a/src/torchattack/dataset.py b/src/torchattack/dataset.py
index 2f21fae..772c5b2 100644
--- a/src/torchattack/dataset.py
+++ b/src/torchattack/dataset.py
@@ -107,7 +107,7 @@ def __init__(
         batch_size: int = 1,
         shuffle: bool = False,
         num_workers: int = 4,
-        normalize: Callable[[torch.Tensor], torch.Tensor] | None = None,
+        transform: Callable[[torch.Tensor], torch.Tensor] | None = None,
         max_samples: int | None = None,
     ):
         # Specifing a custom image root directory is useful when evaluating
@@ -117,7 +117,7 @@ def __init__(
             dataset=NIPSDataset(
                 image_root=image_root if image_root else f'{path}/images',
                 pairs_path=pairs_path if pairs_path else f'{path}/images.csv',
-                transform=normalize,
+                transform=transform,
                 max_samples=max_samples,
             ),
             batch_size=batch_size,
diff --git a/src/torchattack/utils.py b/src/torchattack/utils.py
index 823d08e..8ad2ebe 100644
--- a/src/torchattack/utils.py
+++ b/src/torchattack/utils.py
@@ -32,7 +32,13 @@ def run_attack(attack, attack_cfg, model='resnet50', samples=100, batch_size=8)
     dataloader = NIPSLoader(
         path='data/nips2017',
         batch_size=batch_size,
-        transform=tv.transforms.Resize(size=224, antialias=True),
+        transform=tv.transforms.Compose(
+            [
+                tv.transforms.Resize([232]),
+                tv.transforms.CenterCrop([224]),
+                tv.transforms.ToTensor(),
+            ]
+        ),
         max_samples=samples,
     )
 
@@ -42,7 +48,7 @@ def run_attack(attack, attack_cfg, model='resnet50', samples=100, batch_size=8)
         std=[0.229, 0.224, 0.225],
     )
     total, acc_clean, acc_adv = len(dataloader.dataset), 0, 0  # type: ignore
-    attacker = attack(model=model, transform=normalize, device=device, **attack_cfg)
+    attacker = attack(model=model, normalize=normalize, device=device, **attack_cfg)
     print(attacker)
 
     # Wrap dataloader with rich.progress.track if available