issue1587

albumentations/augmentations/crops/transforms.py

@@ -334,12 +334,11 @@ def apply_to_keypoint(
 
 
 class RandomSizedCrop(_BaseRandomSizedCrop):
-    """Crop a random part of the input and rescale it to some size.
+    """Crop a random portion of the input and rescale it to a specific size.
 
     Args:
         min_max_height ((int, int)): crop size limits.
-        height (int): height after crop and resize.
-        width (int): width after crop and resize.
+        size (tuple[int]): target size for the output image, i.e. (height, width) after crop and resize
         w2h_ratio (float): aspect ratio of crop.
         interpolation (OpenCV flag): flag that is used to specify the interpolation algorithm. Should be one of:
             cv2.INTER_NEAREST, cv2.INTER_LINEAR, cv2.INTER_CUBIC, cv2.INTER_AREA, cv2.INTER_LANCZOS4.
@@ -355,17 +354,49 @@ class RandomSizedCrop(_BaseRandomSizedCrop):
     """
 
     _targets = (Targets.IMAGE, Targets.MASK, Targets.BBOXES, Targets.KEYPOINTS)
+    _size_len = 2
 
     def __init__(
         self,
         min_max_height: Tuple[int, int],
-        height: int,
-        width: int,
+        # NOTE @zetyquickly: when (width, height) are deprecated, make 'size' non optional
+        size: Optional[Union[int, Tuple[int, int]]] = None,
+        width: Optional[int] = None,
+        height: Optional[int] = None,
+        *,
         w2h_ratio: float = 1.0,
         interpolation: int = cv2.INTER_LINEAR,
         always_apply: bool = False,
         p: float = 1.0,
     ):
+        if isinstance(size, tuple):
+            if len(size) != self._size_len:
+                message = "Size must be a tuple of two integers (height, width)."
+                raise ValueError(message)
+            height, width = size
+        elif size is None:
+            if height is None or width is None:
+                message = "If 'size' is not provided, both 'height' and 'width' must be specified."
+                raise ValueError(message)
+            size = (height, width)
+            warn(
+                "Initializing with 'height' and 'width' is deprecated. "
+                "Please use a tuple (height, width) for the 'size' argument.",
+                DeprecationWarning,
+                stacklevel=2,
+            )
+        else:
+            if width is None:
+                message = "When 'size' is an integer, 'width' must be provided."
+                raise ValueError(message)
+            height = size
+            warn(
+                "Initializing with 'size' as an integer and a separate 'width' is deprecated. "
+                "Please use a tuple (height, width) for the 'size' argument.",
+                DeprecationWarning,
+                stacklevel=2,
+            )
+
         super().__init__(height=height, width=width, interpolation=interpolation, always_apply=always_apply, p=p)
 
         self.min_max_height = min_max_height
@@ -388,8 +419,7 @@ class RandomResizedCrop(_BaseRandomSizedCrop):
     """Torchvision's variant of crop a random part of the input and rescale it to some size.
 
     Args:
-        height (int): height after crop and resize.
-        width (int): width after crop and resize.
+        size (tuple[int]): target size for the output image, i.e. (height, width) after crop and resize
         scale ((float, float)): range of size of the origin size cropped
         ratio ((float, float)): range of aspect ratio of the origin aspect ratio cropped
         interpolation (OpenCV flag): flag that is used to specify the interpolation algorithm. Should be one of:
@@ -406,17 +436,49 @@ class RandomResizedCrop(_BaseRandomSizedCrop):
     """
 
     _targets = (Targets.IMAGE, Targets.MASK, Targets.BBOXES, Targets.KEYPOINTS)
+    _size_len = 2
 
     def __init__(
         self,
-        height: int,
-        width: int,
+        # NOTE @zetyquickly: when (width, height) are deprecated, make 'size' non optional
+        size: Optional[Union[int, Tuple[int, int]]] = None,
+        width: Optional[int] = None,
+        height: Optional[int] = None,
+        *,
         scale: Tuple[float, float] = (0.08, 1.0),
         ratio: Tuple[float, float] = (0.75, 1.3333333333333333),
         interpolation: int = cv2.INTER_LINEAR,
         always_apply: bool = False,
         p: float = 1.0,
     ):
+        if isinstance(size, tuple):
+            if len(size) != self._size_len:
+                message = "Size must be a tuple of two integers (height, width)."
+                raise ValueError(message)
+            height, width = size
+        elif size is None:
+            if height is None or width is None:
+                message = "If 'size' is not provided, both 'height' and 'width' must be specified."
+                raise ValueError(message)
+            size = (height, width)
+            warn(
+                "Initializing with 'height' and 'width' is deprecated. "
+                "Please use a tuple (height, width) for the 'size' argument.",
+                DeprecationWarning,
+                stacklevel=2,
+            )
+        else:
+            if width is None:
+                message = "When 'size' is an integer, 'width' must be provided."
+                raise ValueError(message)
+            height = size
+            warn(
+                "Initializing with 'size' as an integer and a separate 'width' is deprecated. "
+                "Please use a tuple (height, width) for the 'size' argument.",
+                DeprecationWarning,
+                stacklevel=2,
+            )
+
         super().__init__(height=height, width=width, interpolation=interpolation, always_apply=always_apply, p=p)
         self.scale = scale
         self.ratio = ratio

tests/test_transforms.py

@@ -4,6 +4,8 @@
 import cv2
 import numpy as np
 import pytest
+import warnings
+from torchvision import transforms as torch_transforms
 
 import albumentations as A
 import albumentations.augmentations.functional as F
@@ -1353,3 +1355,62 @@ def test_spatter_incorrect_color(unsupported_color, mode, message):
         A.Spatter(mode=mode, color=unsupported_color)
 
     assert str(exc_info.value).startswith(message)
+
+@pytest.mark.parametrize("height, width", [(100, 200), (200, 100)])
+@pytest.mark.parametrize("scale", [(0.08, 1.0), (0.5, 1.0)])
+@pytest.mark.parametrize("ratio", [(0.75, 1.33), (1.0, 1.0)])
+def test_random_crop_interfaces_vs_torchvision(height, width, scale, ratio):
+    # NOTE: below will fail when height, width is no longer expected as first two positional arguments
+    transform_albu = A.RandomResizedCrop(height, width, scale=scale, ratio=ratio, p=1)
+    transform_albu_new = A.RandomResizedCrop(size=(height, width), scale=scale, ratio=ratio, p=1)
+
+    image = np.random.randint(0, 256, (224, 224, 3), dtype=np.uint8)
+    transformed_image_albu = transform_albu(image=image)['image']
+    transformed_image_albu_new = transform_albu_new(image=image)['image']
+
+    # PyTorch equivalent operation
+    transform_pt = torch_transforms.RandomResizedCrop(size=(height, width), scale=scale, ratio=ratio)
+    image_pil = torch_transforms.functional.to_pil_image(image)
+    transformed_image_pt = transform_pt(image_pil)
+
+    transformed_image_pt_np = np.array(transformed_image_pt)
+    assert transformed_image_albu.shape == transformed_image_pt_np.shape
+    assert transformed_image_albu_new.shape == transformed_image_pt_np.shape
+
+    # NOTE: below will fail when height, width is no longer expected as second and third positional arguments
+    transform_albu = A.RandomSizedCrop((128, 224), height, width, p=1.0)
+    transform_albu_new = A.RandomSizedCrop(min_max_height=(128, 224), size=(height, width), p=1.0)
+    transformed_image_albu = transform_albu(image=image)['image']
+    transformed_image_albu_new = transform_albu_new(image=image)['image']
+    assert transformed_image_albu.shape == transformed_image_pt_np.shape
+    assert transformed_image_albu_new.shape == transformed_image_pt_np.shape
+
+    # NOTE: below will fail when height, width is no longer expected as first two positional arguments
+    transform_albu = A.RandomResizedCrop(height, width, scale=scale, ratio=ratio, p=1)
+    transform_albu_height_is_size = A.RandomResizedCrop(size=height, width=width, scale=scale, ratio=ratio, p=1)
+
+    image = np.random.randint(0, 256, (224, 224, 3), dtype=np.uint8)
+    transformed_image_albu = transform_albu(image=image)['image']
+    transform_albu_height_is_size = transform_albu_new(image=image)['image']
+    assert transformed_image_albu.shape == transformed_image_pt_np.shape
+    assert transform_albu_height_is_size.shape == transformed_image_pt_np.shape
+
+@pytest.mark.parametrize("size, width, height, expected_warning", [
+    ((100, 200), None, None, None),
+    (None, 200, 100, DeprecationWarning),
+    (100, None, None, ValueError),
+])
+def test_deprecation_warnings(size, width, height, expected_warning):
+    with warnings.catch_warnings(record=True) as w:
+        warnings.simplefilter("always")
+        if expected_warning == ValueError:
+            with pytest.raises(ValueError):
+                A.RandomResizedCrop(size=size, width=width, height=height)
+        else:
+            A.RandomResizedCrop(size=size, width=width, height=height)
+        if expected_warning is DeprecationWarning:
+            assert len(w) == 1
+            assert issubclass(w[-1].category, expected_warning)
+        else:
+            assert not w
+    warnings.resetwarnings()