Expand prototype transforms consistency tests to most random tests (#6522)

* add checks for pure random apply transformations

* add tests for some non-deterministic transforms used in classification reference

* add consistency checks for most transforms

* fix RandomAffine, RandomPerspective, and GaussianBlur

* fix tests

test/test_prototype_transforms.py

@@ -555,10 +555,10 @@ class TestRandomAffine:
         if translate is not None:
             w_max = int(round(translate[0] * w))
             h_max = int(round(translate[1] * h))
-            assert -w_max <= params["translations"][0] <= w_max
-            assert -h_max <= params["translations"][1] <= h_max
+            assert -w_max <= params["translate"][0] <= w_max
+            assert -h_max <= params["translate"][1] <= h_max
         else:
-            assert params["translations"] == (0, 0)
+            assert params["translate"] == (0, 0)
 
         if scale is not None:
             assert scale[0] <= params["scale"] <= scale[1]
@@ -759,7 +759,8 @@ class TestGaussianBlur:
         if isinstance(kernel_size, (tuple, list)):
             assert transform.kernel_size == kernel_size
         else:
-            assert transform.kernel_size == (kernel_size, kernel_size)
+            kernel_size = (kernel_size, kernel_size)
+            assert transform.kernel_size == kernel_size
 
         if isinstance(sigma, (tuple, list)):
             assert transform.sigma == sigma
@@ -779,7 +780,7 @@ class TestGaussianBlur:
         torch.manual_seed(12)
         params = transform._get_params(inpt)
 
-        fn.assert_called_once_with(inpt, **params)
+        fn.assert_called_once_with(inpt, kernel_size, **params)
 
 
 class TestRandomColorOp:

test/test_prototype_transforms_consistency.py

@@ -137,6 +137,7 @@ CONSISTENCY_CONFIGS = [
         [
             ArgsKwargs(18),
             ArgsKwargs((18, 13)),
+            ArgsKwargs(18, vertical_flip=True),
         ],
         make_images_kwargs=dict(DEFAULT_MAKE_IMAGES_KWARGS, sizes=[(20, 19)]),
     ),
@@ -215,6 +216,214 @@ CONSISTENCY_CONFIGS = [
         # images given that the transform does nothing but call it anyway.
         supports_pil=False,
     ),
+    ConsistencyConfig(
+        prototype_transforms.RandomHorizontalFlip,
+        legacy_transforms.RandomHorizontalFlip,
+        [
+            ArgsKwargs(p=0),
+            ArgsKwargs(p=1),
+        ],
+    ),
+    ConsistencyConfig(
+        prototype_transforms.RandomVerticalFlip,
+        legacy_transforms.RandomVerticalFlip,
+        [
+            ArgsKwargs(p=0),
+            ArgsKwargs(p=1),
+        ],
+    ),
+    ConsistencyConfig(
+        prototype_transforms.RandomEqualize,
+        legacy_transforms.RandomEqualize,
+        [
+            ArgsKwargs(p=0),
+            ArgsKwargs(p=1),
+        ],
+        make_images_kwargs=dict(DEFAULT_MAKE_IMAGES_KWARGS, dtypes=[torch.uint8]),
+    ),
+    ConsistencyConfig(
+        prototype_transforms.RandomInvert,
+        legacy_transforms.RandomInvert,
+        [
+            ArgsKwargs(p=0),
+            ArgsKwargs(p=1),
+        ],
+    ),
+    ConsistencyConfig(
+        prototype_transforms.RandomPosterize,
+        legacy_transforms.RandomPosterize,
+        [
+            ArgsKwargs(p=0, bits=5),
+            ArgsKwargs(p=1, bits=1),
+            ArgsKwargs(p=1, bits=3),
+        ],
+        make_images_kwargs=dict(DEFAULT_MAKE_IMAGES_KWARGS, dtypes=[torch.uint8]),
+    ),
+    ConsistencyConfig(
+        prototype_transforms.RandomSolarize,
+        legacy_transforms.RandomSolarize,
+        [
+            ArgsKwargs(p=0, threshold=0.5),
+            ArgsKwargs(p=1, threshold=0.3),
+            ArgsKwargs(p=1, threshold=0.99),
+        ],
+    ),
+    ConsistencyConfig(
+        prototype_transforms.RandomAutocontrast,
+        legacy_transforms.RandomAutocontrast,
+        [
+            ArgsKwargs(p=0),
+            ArgsKwargs(p=1),
+        ],
+    ),
+    ConsistencyConfig(
+        prototype_transforms.RandomAdjustSharpness,
+        legacy_transforms.RandomAdjustSharpness,
+        [
+            ArgsKwargs(p=0, sharpness_factor=0.5),
+            ArgsKwargs(p=1, sharpness_factor=0.3),
+            ArgsKwargs(p=1, sharpness_factor=0.99),
+        ],
+    ),
+    ConsistencyConfig(
+        prototype_transforms.RandomGrayscale,
+        legacy_transforms.RandomGrayscale,
+        [
+            ArgsKwargs(p=0),
+            ArgsKwargs(p=1),
+        ],
+    ),
+    ConsistencyConfig(
+        prototype_transforms.RandomResizedCrop,
+        legacy_transforms.RandomResizedCrop,
+        [
+            ArgsKwargs(16),
+            ArgsKwargs(17, scale=(0.3, 0.7)),
+            ArgsKwargs(25, ratio=(0.5, 1.5)),
+            ArgsKwargs((31, 28), interpolation=prototype_transforms.InterpolationMode.NEAREST),
+            ArgsKwargs((33, 26), interpolation=prototype_transforms.InterpolationMode.BICUBIC),
+            ArgsKwargs((29, 32), antialias=False),
+            ArgsKwargs((28, 31), antialias=True),
+        ],
+    ),
+    ConsistencyConfig(
+        prototype_transforms.RandomErasing,
+        legacy_transforms.RandomErasing,
+        [
+            ArgsKwargs(p=0),
+            ArgsKwargs(p=1),
+            ArgsKwargs(p=1, scale=(0.3, 0.7)),
+            ArgsKwargs(p=1, ratio=(0.5, 1.5)),
+            ArgsKwargs(p=1, value=1),
+            ArgsKwargs(p=1, value=(1, 2, 3)),
+            ArgsKwargs(p=1, value="random"),
+        ],
+        supports_pil=False,
+    ),
+    ConsistencyConfig(
+        prototype_transforms.ColorJitter,
+        legacy_transforms.ColorJitter,
+        [
+            ArgsKwargs(),
+            ArgsKwargs(brightness=0.1),
+            ArgsKwargs(brightness=(0.2, 0.3)),
+            ArgsKwargs(contrast=0.4),
+            ArgsKwargs(contrast=(0.5, 0.6)),
+            ArgsKwargs(saturation=0.7),
+            ArgsKwargs(saturation=(0.8, 0.9)),
+            ArgsKwargs(hue=0.3),
+            ArgsKwargs(hue=(-0.1, 0.2)),
+            ArgsKwargs(brightness=0.1, contrast=0.4, saturation=0.7, hue=0.3),
+        ],
+    ),
+    ConsistencyConfig(
+        prototype_transforms.ElasticTransform,
+        legacy_transforms.ElasticTransform,
+        [
+            ArgsKwargs(),
+            ArgsKwargs(alpha=20.0),
+            ArgsKwargs(alpha=(15.3, 27.2)),
+            ArgsKwargs(sigma=3.0),
+            ArgsKwargs(sigma=(2.5, 3.9)),
+            ArgsKwargs(interpolation=prototype_transforms.InterpolationMode.NEAREST),
+            ArgsKwargs(interpolation=prototype_transforms.InterpolationMode.BICUBIC),
+            ArgsKwargs(fill=1),
+        ],
+        # ElasticTransform needs larger images to avoid the needed internal padding being larger than the actual image
+        make_images_kwargs=dict(DEFAULT_MAKE_IMAGES_KWARGS, sizes=[(163, 163), (72, 333), (313, 95)]),
+    ),
+    ConsistencyConfig(
+        prototype_transforms.GaussianBlur,
+        legacy_transforms.GaussianBlur,
+        [
+            ArgsKwargs(kernel_size=3),
+            ArgsKwargs(kernel_size=(1, 5)),
+            ArgsKwargs(kernel_size=3, sigma=0.7),
+            ArgsKwargs(kernel_size=5, sigma=(0.3, 1.4)),
+        ],
+    ),
+    ConsistencyConfig(
+        prototype_transforms.RandomAffine,
+        legacy_transforms.RandomAffine,
+        [
+            ArgsKwargs(degrees=30.0),
+            ArgsKwargs(degrees=(-20.0, 10.0)),
+            ArgsKwargs(degrees=0.0, translate=(0.4, 0.6)),
+            ArgsKwargs(degrees=0.0, scale=(0.3, 0.8)),
+            ArgsKwargs(degrees=0.0, shear=13),
+            ArgsKwargs(degrees=0.0, shear=(8, 17)),
+            ArgsKwargs(degrees=0.0, shear=(4, 5, 4, 13)),
+            ArgsKwargs(degrees=(-20.0, 10.0), translate=(0.4, 0.6), scale=(0.3, 0.8), shear=(4, 5, 4, 13)),
+            ArgsKwargs(degrees=30.0, interpolation=prototype_transforms.InterpolationMode.NEAREST),
+            ArgsKwargs(degrees=30.0, fill=1),
+            ArgsKwargs(degrees=30.0, fill=(2, 3, 4)),
+            ArgsKwargs(degrees=30.0, center=(0, 0)),
+        ],
+    ),
+    ConsistencyConfig(
+        prototype_transforms.RandomCrop,
+        legacy_transforms.RandomCrop,
+        [
+            ArgsKwargs(12),
+            ArgsKwargs((15, 17)),
+            ArgsKwargs(11, padding=1),
+            ArgsKwargs((8, 13), padding=(2, 3)),
+            ArgsKwargs((14, 9), padding=(0, 2, 1, 0)),
+            ArgsKwargs(36, pad_if_needed=True),
+            ArgsKwargs((7, 8), fill=1),
+            ArgsKwargs(5, fill=(1, 2, 3)),
+            ArgsKwargs(12),
+            ArgsKwargs(15, padding=2, padding_mode="edge"),
+            ArgsKwargs(17, padding=(1, 0), padding_mode="reflect"),
+            ArgsKwargs(8, padding=(3, 0, 0, 1), padding_mode="symmetric"),
+        ],
+        make_images_kwargs=dict(DEFAULT_MAKE_IMAGES_KWARGS, sizes=[(26, 26), (18, 33), (29, 22)]),
+    ),
+    ConsistencyConfig(
+        prototype_transforms.RandomPerspective,
+        legacy_transforms.RandomPerspective,
+        [
+            ArgsKwargs(p=0),
+            ArgsKwargs(p=1),
+            ArgsKwargs(p=1, distortion_scale=0.3),
+            ArgsKwargs(p=1, distortion_scale=0.2, interpolation=prototype_transforms.InterpolationMode.NEAREST),
+            ArgsKwargs(p=1, distortion_scale=0.1, fill=1),
+            ArgsKwargs(p=1, distortion_scale=0.4, fill=(1, 2, 3)),
+        ],
+    ),
+    ConsistencyConfig(
+        prototype_transforms.RandomRotation,
+        legacy_transforms.RandomRotation,
+        [
+            ArgsKwargs(degrees=30.0),
+            ArgsKwargs(degrees=(-20.0, 10.0)),
+            ArgsKwargs(degrees=30.0, interpolation=prototype_transforms.InterpolationMode.BILINEAR),
+            ArgsKwargs(degrees=30.0, expand=True),
+            ArgsKwargs(degrees=30.0, center=(0, 0)),
+            ArgsKwargs(degrees=30.0, fill=1),
+            ArgsKwargs(degrees=30.0, fill=(1, 2, 3)),
+        ],
+    ),
 ]
 
 
@@ -227,22 +436,22 @@ def test_automatic_coverage_deterministic():
         and not issubclass(obj, enum.Enum)
         and name
         not in {
-            "Compose",
             # This framework is based on the assumption that the input image can always be a tensor and optionally a
-            # PIL image. The transforms below require a non-tensor input and thus have to be tested manually.
+            # PIL image, but the transforms below require a non-tensor input.
             "PILToTensor",
             "ToTensor",
-        }
-    }
-    # filter out random transformations
-    legacy = {name for name in legacy if "Random" not in name} - {
+            # Transform containers cannot be tested without other tranforms
+            "Compose",
+            "RandomApply",
+            "RandomChoice",
+            "RandomOrder",
+            # If the random parameter generation in the legacy and prototype transform is the same, setting the seed
+            # should be sufficient. In that case, the transforms below should be tested automatically.
             "AugMix",
-        "TrivialAugmentWide",
-        "GaussianBlur",
-        "RandAugment",
             "AutoAugment",
-        "ColorJitter",
-        "ElasticTransform",
+            "RandAugment",
+            "TrivialAugmentWide",
+        }
     }
 
     prototype = {config.legacy_cls.__name__ for config in CONSISTENCY_CONFIGS}
@@ -285,6 +494,7 @@ def test_consistency(prototype_transform_cls, legacy_transform_cls, args_kwargs,
         image_repr = f"[{tuple(image.shape)}, {str(image.dtype).rsplit('.')[-1]}]"
 
         try:
+            torch.manual_seed(0)
             output_legacy_tensor = legacy(image_tensor)
         except Exception as exc:
             raise pytest.UsageError(
@@ -294,6 +504,7 @@ def test_consistency(prototype_transform_cls, legacy_transform_cls, args_kwargs,
             ) from exc
 
         try:
+            torch.manual_seed(0)
             output_prototype_tensor = prototype(image_tensor)
         except Exception as exc:
             raise AssertionError(
@@ -309,6 +520,7 @@ def test_consistency(prototype_transform_cls, legacy_transform_cls, args_kwargs,
         )
 
         try:
+            torch.manual_seed(0)
             output_prototype_image = prototype(image)
         except Exception as exc:
             raise AssertionError(
@@ -325,6 +537,7 @@ def test_consistency(prototype_transform_cls, legacy_transform_cls, args_kwargs,
 
         if image_pil is not None:
             try:
+                torch.manual_seed(0)
                 output_legacy_pil = legacy(image_pil)
             except Exception as exc:
                 raise pytest.UsageError(
@@ -334,6 +547,7 @@ def test_consistency(prototype_transform_cls, legacy_transform_cls, args_kwargs,
                 ) from exc
 
             try:
+                torch.manual_seed(0)
                 output_prototype_pil = prototype(image_pil)
             except Exception as exc:
                 raise AssertionError(

torchvision/prototype/transforms/_geometry.py

@@ -373,9 +373,9 @@ class RandomAffine(Transform):
             max_dy = float(self.translate[1] * height)
             tx = int(round(torch.empty(1).uniform_(-max_dx, max_dx).item()))
             ty = int(round(torch.empty(1).uniform_(-max_dy, max_dy).item()))
-            translations = (tx, ty)
+            translate = (tx, ty)
         else:
-            translations = (0, 0)
+            translate = (0, 0)
 
         if self.scale is not None:
             scale = float(torch.empty(1).uniform_(self.scale[0], self.scale[1]).item())
@@ -389,7 +389,7 @@ class RandomAffine(Transform):
                 shear_y = float(torch.empty(1).uniform_(self.shear[2], self.shear[3]).item())
 
         shear = (shear_x, shear_y)
-        return dict(angle=angle, translations=translations, scale=scale, shear=shear)
+        return dict(angle=angle, translate=translate, scale=scale, shear=shear)
 
     def _transform(self, inpt: Any, params: Dict[str, Any]) -> Any:
         return F.affine(
@@ -489,7 +489,7 @@ class RandomCrop(Transform):
 class RandomPerspective(_RandomApplyTransform):
     def __init__(
         self,
-        distortion_scale: float,
+        distortion_scale: float = 0.5,
         fill: Union[int, float, Sequence[int], Sequence[float]] = 0,
         interpolation: InterpolationMode = InterpolationMode.BILINEAR,
         p: float = 0.5,

torchvision/prototype/transforms/_misc.py

@@ -136,7 +136,7 @@ class GaussianBlur(Transform):
         return dict(sigma=[sigma, sigma])
 
     def _transform(self, inpt: Any, params: Dict[str, Any]) -> Any:
-        return F.gaussian_blur(inpt, **params)
+        return F.gaussian_blur(inpt, self.kernel_size, **params)
 
 
 class ToDtype(Lambda):