feat: use resize transform

src/augmentations.py

@@ -7,6 +7,17 @@ from timm.data.constants import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
 from torchvision import transforms
 
 
+class GroupResize:
+    def __init__(self, size: int = 256) -> None:
+        self.transform = transforms.Resize(size)
+
+    def __call__(
+        self, img_tuple: Tuple[torch.Tensor, torch.Tensor]
+    ) -> Tuple[List[torch.Tensor], torch.Tensor]:
+        img_group, label = img_tuple
+        return [self.transform(img) for img in img_group], label
+
+
 class GroupNormalize:
     def __init__(self, mean: List[float], std: List[float]) -> None:
         self.mean = mean
@@ -109,6 +120,7 @@ class TubeMaskingGenerator:
 def get_videomae_transform(input_size: int = 224) -> "transforms.Compose":
     return transforms.Compose(
         [
+            GroupResize(size=384),
             GroupCenterCrop(input_size),
             Stack(roll=False),
             ToTorchFormatTensor(div=True),