Added decorators, Updated UA code and description

updated UA parameter check, fixed PyLint errors.

fixed UA parameter validator bug

Updated UA C++ description

Changed py and cpp UA decorator names

mindspore/dataset/transforms/vision/c_transforms.py

@@ -45,7 +45,7 @@ import mindspore._c_dataengine as cde
 from .utils import Inter, Border
 from .validators import check_prob, check_crop, check_resize_interpolation, check_random_resize_crop, \
     check_normalize_c, check_random_crop, check_random_color_adjust, check_random_rotation, \
-    check_resize, check_rescale, check_pad, check_cutout, check_uniform_augmentation
+    check_resize, check_rescale, check_pad, check_cutout, check_uniform_augment_cpp
 
 DE_C_INTER_MODE = {Inter.NEAREST: cde.InterpolationMode.DE_INTER_NEAREST_NEIGHBOUR,
                    Inter.LINEAR: cde.InterpolationMode.DE_INTER_LINEAR,
@@ -456,7 +456,7 @@ class UniformAugment(cde.UniformAugOp):
 
     Args:
         operations: list of C++ operations (python OPs are not accepted).
-        num_ops (int): number of OPs to be selected and applied.
+        num_ops (int, optional): number of OPs to be selected and applied (default=2).
 
     Examples:
         >>> transforms_list = [c_transforms.RandomHorizontalFlip(),
@@ -470,7 +470,7 @@ class UniformAugment(cde.UniformAugOp):
         >>>                operations=transforms_all, num_parallel_workers=1)
     """
 
-    @check_uniform_augmentation
+    @check_uniform_augment_cpp
     def __init__(self, operations, num_ops=2):
         self.operations = operations
         self.num_ops = num_ops

mindspore/dataset/transforms/vision/py_transforms.py

@@ -32,7 +32,7 @@ from .validators import check_prob, check_crop, check_resize_interpolation, chec
     check_normalize_py, check_random_crop, check_random_color_adjust, check_random_rotation, \
     check_transforms_list, check_random_apply, check_ten_crop, check_num_channels, check_pad, \
     check_random_perspective, check_random_erasing, check_cutout, check_linear_transform, check_random_affine, \
-    check_mix_up
+    check_mix_up, check_positive_degrees, check_uniform_augment_py
 from .utils import Inter, Border
 
 DE_PY_INTER_MODE = {Inter.NEAREST: Image.NEAREST,
@@ -1329,6 +1329,7 @@ class RandomColor:
         >>>                          py_transforms.ToTensor()])
     """
 
+    @check_positive_degrees
     def __init__(self, degrees=(0.1, 1.9)):
         self.degrees = degrees
 
@@ -1361,6 +1362,7 @@ class RandomSharpness:
 
     """
 
+    @check_positive_degrees
     def __init__(self, degrees=(0.1, 1.9)):
         self.degrees = degrees
 
@@ -1458,6 +1460,7 @@ class UniformAugment:
     Uniformly select and apply a number of transforms sequentially from
     a list of transforms. Randomly assigns a probability to each transform for
     each image to decide whether apply it or not.
+    All the transforms in transform list must have the same input/output data type.
 
     Args:
          transforms (list): List of transformations to be chosen from to apply.
@@ -1473,6 +1476,7 @@ class UniformAugment:
         >>>                          py_transforms.ToTensor()])
     """
 
+    @check_uniform_augment_py
     def __init__(self, transforms, num_ops=2):
         self.transforms = transforms
         self.num_ops = num_ops

mindspore/dataset/transforms/vision/py_transforms_util.py

@@ -1427,17 +1427,6 @@ def random_color(img, degrees):
     if not is_pil(img):
         raise TypeError('img should be PIL Image. Got {}'.format(type(img)))
 
-    if isinstance(degrees, (list, tuple)):
-        if len(degrees) != 2:
-            raise ValueError("Degrees must be a sequence length 2.")
-        if degrees[0] < 0:
-            raise ValueError("Degree value must be non-negative.")
-        if degrees[0] > degrees[1]:
-            raise ValueError("Degrees should be in (min,max) format. Got (max,min).")
-
-    else:
-        raise TypeError("Degrees must be a sequence in (min,max) format.")
-
     v = (degrees[1] - degrees[0]) * random.random() + degrees[0]
     return ImageEnhance.Color(img).enhance(v)
 
@@ -1459,17 +1448,6 @@ def random_sharpness(img, degrees):
     if not is_pil(img):
         raise TypeError('img should be PIL Image. Got {}'.format(type(img)))
 
-    if isinstance(degrees, (list, tuple)):
-        if len(degrees) != 2:
-            raise ValueError("Degrees must be a sequence length 2.")
-        if degrees[0] < 0:
-            raise ValueError("Degree value must be non-negative.")
-        if degrees[0] > degrees[1]:
-            raise ValueError("Degrees should be in (min,max) format. Got (max,min).")
-
-    else:
-        raise TypeError("Degrees must be a sequence in (min,max) format.")
-
     v = (degrees[1] - degrees[0]) * random.random() + degrees[0]
     return ImageEnhance.Sharpness(img).enhance(v)
 
@@ -1537,6 +1515,7 @@ def uniform_augment(img, transforms, num_ops):
     Uniformly select and apply a number of transforms sequentially from
     a list of transforms. Randomly assigns a probability to each transform for
     each image to decide whether apply it or not.
+    All the transforms in transform list must have the same input/output data type.
 
     Args:
         img: Image to be applied transformation.
@@ -1545,23 +1524,14 @@ def uniform_augment(img, transforms, num_ops):
 
     Returns:
         img, Transformed image.
-    """
 
-    if transforms is None:
-        raise ValueError("transforms is not provided.")
-    if not isinstance(transforms, list):
-        raise ValueError("The transforms needs to be a list.")
-
-    if not isinstance(num_ops, int):
-        raise ValueError("Number of operations should be a positive integer.")
-    if num_ops < 1:
-        raise ValueError("Number of operators should equal or greater than one.")
+    """
 
-    for _ in range(num_ops):
-        AugmentOp = random.choice(transforms)
+    op_idx = np.random.choice(len(transforms), size=num_ops, replace=False)
+    for idx in op_idx:
+        AugmentOp = transforms[idx]
         pr = random.random()
         if random.random() < pr:
             img = AugmentOp(img.copy())
-        transforms.remove(AugmentOp)
 
     return img

mindspore/dataset/transforms/vision/validators.py

@@ -816,8 +816,8 @@ def check_rescale(method):
     return new_method
 
 
-def check_uniform_augmentation(method):
-    """Wrapper method to check the parameters of UniformAugmentation."""
+def check_uniform_augment_cpp(method):
+    """Wrapper method to check the parameters of UniformAugment cpp op."""
 
     @wraps(method)
     def new_method(self, *args, **kwargs):
@@ -850,3 +850,60 @@ def check_uniform_augmentation(method):
         return method(self, **kwargs)
 
     return new_method
+
+
+def check_uniform_augment_py(method):
+    """Wrapper method to check the parameters of python UniformAugment op."""
+
+    @wraps(method)
+    def new_method(self, *args, **kwargs):
+        transforms, num_ops = (list(args) + 2 * [None])[:2]
+        if "transforms" in kwargs:
+            transforms = kwargs.get("transforms")
+        if transforms is None:
+            raise ValueError("transforms is not provided.")
+        if not transforms:
+            raise ValueError("transforms list is empty.")
+        check_list(transforms)
+        for transform in transforms:
+            if isinstance(transform, TensorOp):
+                raise ValueError("transform list only accepts Python operations.")
+        kwargs["transforms"] = transforms
+
+        if "num_ops" in kwargs:
+            num_ops = kwargs.get("num_ops")
+        if num_ops is not None:
+            check_type(num_ops, int)
+            check_positive(num_ops)
+            if num_ops > len(transforms):
+                raise ValueError("num_ops cannot be greater than the length of transforms list.")
+            kwargs["num_ops"] = num_ops
+
+        return method(self, **kwargs)
+
+    return new_method
+
+
+def check_positive_degrees(method):
+    """A wrapper method to check degrees parameter in RandSharpness and RandColor"""
+
+    @wraps(method)
+    def new_method(self, *args, **kwargs):
+        degrees = (list(args) + [None])[0]
+        if "degrees" in kwargs:
+            degrees = kwargs.get("degrees")
+
+        if degrees is not None:
+            if isinstance(degrees, (list, tuple)):
+                if len(degrees) != 2:
+                    raise ValueError("Degrees must be a sequence with length 2.")
+                if degrees[0] < 0:
+                    raise ValueError("Degrees range must be non-negative.")
+                if degrees[0] > degrees[1]:
+                    raise ValueError("Degrees should be in (min,max) format. Got (max,min).")
+            else:
+                raise TypeError("Degrees must be a sequence in (min,max) format.")
+
+        return method(self, **kwargs)
+
+    return new_method