More cleanup for prototype transforms (#6500)

* add aliases for hflip and vflip

* reduce imports from torchvision.transforms in torchvision.prototype.transforms

* add aliases for to_pil_image abd pil_to_tensor

* deprecate to_tensor

* add some FIXME cleanup comments

* address reviews

* add dimension getters

* undeprecate PILToTensor and ToPILImage

* address review

* fix test

test/test_prototype_transforms.py

@@ -1071,10 +1071,9 @@ class TestToPILImage:
         [torch.Tensor, PIL.Image.Image, features.Image, np.ndarray, features.BoundingBox, str, int],
     )
     def test__transform(self, inpt_type, mocker):
-        fn = mocker.patch("torchvision.transforms.functional.to_pil_image")
+        fn = mocker.patch("torchvision.prototype.transforms.functional.to_image_pil")
 
         inpt = mocker.MagicMock(spec=inpt_type)
-        with pytest.warns(UserWarning, match="deprecated and will be removed"):
         transform = transforms.ToPILImage()
         transform(inpt)
         if inpt_type in (PIL.Image.Image, features.BoundingBox, str, int):

test/test_prototype_transforms_functional.py

@@ -674,6 +674,8 @@ def erase_image_tensor():
         and name
         not in {
             "to_image_tensor",
+            "get_image_num_channels",
+            "get_image_size",
         }
     ],
 )

torchvision/prototype/transforms/__init__.py

+from torchvision.transforms import AutoAugmentPolicy, InterpolationMode  # usort: skip
+
 from . import functional  # usort: skip
 
 from ._transform import Transform  # usort: skip
 
 from ._augment import RandomCutmix, RandomErasing, RandomMixup, SimpleCopyPaste
-from ._auto_augment import AugMix, AutoAugment, AutoAugmentPolicy, RandAugment, TrivialAugmentWide
+from ._auto_augment import AugMix, AutoAugment, RandAugment, TrivialAugmentWide
 from ._color import (
     ColorJitter,
     RandomAdjustSharpness,
@@ -37,6 +39,6 @@ from ._geometry import (
 )
 from ._meta import ClampBoundingBoxes, ConvertBoundingBoxFormat, ConvertColorSpace, ConvertImageDtype
 from ._misc import GaussianBlur, Identity, Lambda, LinearTransformation, Normalize, RemoveSmallBoundingBoxes, ToDtype
-from ._type_conversion import DecodeImage, LabelToOneHot, ToImagePIL, ToImageTensor
+from ._type_conversion import DecodeImage, LabelToOneHot, PILToTensor, ToImagePIL, ToImageTensor, ToPILImage
 
-from ._deprecated import Grayscale, RandomGrayscale, ToTensor, ToPILImage, PILToTensor  # usort: skip
+from ._deprecated import Grayscale, RandomGrayscale, ToTensor  # usort: skip

torchvision/prototype/transforms/_augment.py

@@ -8,9 +8,7 @@ import torch
 from torch.utils._pytree import tree_flatten, tree_unflatten
 from torchvision.ops import masks_to_boxes
 from torchvision.prototype import features
-
-from torchvision.prototype.transforms import functional as F
-from torchvision.transforms.functional import InterpolationMode, pil_to_tensor
+from torchvision.prototype.transforms import functional as F, InterpolationMode
 
 from ._transform import _RandomApplyTransform
 from ._utils import has_any, query_chw
@@ -279,7 +277,7 @@ class SimpleCopyPaste(_RandomApplyTransform):
             if isinstance(obj, features.Image) or features.is_simple_tensor(obj):
                 images.append(obj)
             elif isinstance(obj, PIL.Image.Image):
-                images.append(pil_to_tensor(obj))
+                images.append(F.to_image_tensor(obj))
             elif isinstance(obj, features.BoundingBox):
                 bboxes.append(obj)
             elif isinstance(obj, features.SegmentationMask):

torchvision/prototype/transforms/_auto_augment.py

@@ -7,11 +7,10 @@ import torch
 
 from torch.utils._pytree import tree_flatten, tree_unflatten
 from torchvision.prototype import features
-from torchvision.prototype.transforms import functional as F, Transform
-from torchvision.transforms.autoaugment import AutoAugmentPolicy
-from torchvision.transforms.functional import InterpolationMode, pil_to_tensor, to_pil_image
+from torchvision.prototype.transforms import AutoAugmentPolicy, functional as F, InterpolationMode, Transform
+from torchvision.prototype.transforms.functional._meta import get_chw
 
-from ._utils import _isinstance, get_chw
+from ._utils import _isinstance
 
 K = TypeVar("K")
 V = TypeVar("V")
@@ -473,7 +472,7 @@ class AugMix(_AutoAugmentBase):
         if isinstance(orig_image, torch.Tensor):
             image = orig_image
         else:  # isinstance(inpt, PIL.Image.Image):
-            image = pil_to_tensor(orig_image)
+            image = F.to_image_tensor(orig_image)
 
         augmentation_space = self._AUGMENTATION_SPACE if self.all_ops else self._PARTIAL_AUGMENTATION_SPACE
 
@@ -516,6 +515,6 @@ class AugMix(_AutoAugmentBase):
         if isinstance(orig_image, features.Image):
             mix = features.Image.new_like(orig_image, mix)
         elif isinstance(orig_image, PIL.Image.Image):
-            mix = to_pil_image(mix)
+            mix = F.to_image_pil(mix)
 
         return self._put_into_sample(sample, id, mix)

torchvision/prototype/transforms/_color.py

@@ -5,7 +5,6 @@ import PIL.Image
 import torch
 from torchvision.prototype import features
 from torchvision.prototype.transforms import functional as F, Transform
-from torchvision.transforms import functional as _F
 
 from ._transform import _RandomApplyTransform
 from ._utils import query_chw
@@ -85,6 +84,8 @@ class ColorJitter(Transform):
 
 
 class RandomPhotometricDistort(Transform):
+    _transformed_types = (features.Image, PIL.Image.Image, features.is_simple_tensor)
+
     def __init__(
         self,
         contrast: Tuple[float, float] = (0.5, 1.5),
@@ -112,19 +113,15 @@ class RandomPhotometricDistort(Transform):
         )
 
     def _permute_channels(self, inpt: Any, *, permutation: torch.Tensor) -> Any:
-        if not (isinstance(inpt, (features.Image, PIL.Image.Image)) or features.is_simple_tensor(inpt)):
-            return inpt
-
-        image = inpt
         if isinstance(inpt, PIL.Image.Image):
-            image = _F.pil_to_tensor(image)
+            inpt = F.to_image_tensor(inpt)
 
-        output = image[..., permutation, :, :]
+        output = inpt[..., permutation, :, :]
 
         if isinstance(inpt, features.Image):
             output = features.Image.new_like(inpt, output, color_space=features.ColorSpace.OTHER)
         elif isinstance(inpt, PIL.Image.Image):
-            output = _F.to_pil_image(output)
+            output = F.to_image_pil(output)
 
         return output
 

torchvision/prototype/transforms/_deprecated.py

 import warnings
-from typing import Any, Dict, Optional
+from typing import Any, Dict
 
 import numpy as np
 import PIL.Image
@@ -20,7 +20,7 @@ class ToTensor(Transform):
     def __init__(self) -> None:
         warnings.warn(
             "The transform `ToTensor()` is deprecated and will be removed in a future release. "
-            "Instead, please use `transforms.ToImageTensor()`."
+            "Instead, please use `transforms.Compose([transforms.ToImageTensor(), transforms.ConvertImageDtype()])`."
         )
         super().__init__()
 
@@ -28,35 +28,6 @@ class ToTensor(Transform):
         return _F.to_tensor(inpt)
 
 
-class PILToTensor(Transform):
-    _transformed_types = (PIL.Image.Image,)
-
-    def __init__(self) -> None:
-        warnings.warn(
-            "The transform `PILToTensor()` is deprecated and will be removed in a future release. "
-            "Instead, please use `transforms.ToImageTensor()`."
-        )
-        super().__init__()
-
-    def _transform(self, inpt: Any, params: Dict[str, Any]) -> torch.Tensor:
-        return _F.pil_to_tensor(inpt)
-
-
-class ToPILImage(Transform):
-    _transformed_types = (features.is_simple_tensor, features.Image, np.ndarray)
-
-    def __init__(self, mode: Optional[str] = None) -> None:
-        warnings.warn(
-            "The transform `ToPILImage()` is deprecated and will be removed in a future release. "
-            "Instead, please use `transforms.ToImagePIL()`."
-        )
-        super().__init__()
-        self.mode = mode
-
-    def _transform(self, inpt: Any, params: Dict[str, Any]) -> PIL.Image.Image:
-        return _F.to_pil_image(inpt, mode=self.mode)
-
-
 class Grayscale(Transform):
     _transformed_types = (features.Image, PIL.Image.Image, features.is_simple_tensor)
 

torchvision/prototype/transforms/_geometry.py

@@ -7,15 +7,21 @@ import PIL.Image
 import torch
 from torchvision.ops.boxes import box_iou
 from torchvision.prototype import features
-from torchvision.prototype.transforms import functional as F, Transform
-from torchvision.transforms.functional import InterpolationMode
-from torchvision.transforms.functional_tensor import _parse_pad_padding
-from torchvision.transforms.transforms import _check_sequence_input, _setup_angle, _setup_size
+from torchvision.prototype.transforms import functional as F, InterpolationMode, Transform
 
 from typing_extensions import Literal
 
 from ._transform import _RandomApplyTransform
-from ._utils import has_all, has_any, query_bounding_box, query_chw
+from ._utils import (
+    _check_sequence_input,
+    _parse_pad_padding,
+    _setup_angle,
+    _setup_size,
+    has_all,
+    has_any,
+    query_bounding_box,
+    query_chw,
+)
 
 
 class RandomHorizontalFlip(_RandomApplyTransform):

torchvision/prototype/transforms/_meta.py

@@ -5,7 +5,6 @@ import PIL.Image
 import torch
 from torchvision.prototype import features
 from torchvision.prototype.transforms import functional as F, Transform
-from torchvision.transforms.functional import convert_image_dtype
 
 
 class ConvertBoundingBoxFormat(Transform):
@@ -30,7 +29,7 @@ class ConvertImageDtype(Transform):
         self.dtype = dtype
 
     def _transform(self, inpt: Any, params: Dict[str, Any]) -> Any:
-        output = convert_image_dtype(inpt, dtype=self.dtype)
+        output = F.convert_image_dtype(inpt, dtype=self.dtype)
         return output if features.is_simple_tensor(inpt) else features.Image.new_like(inpt, output, dtype=self.dtype)
 
 

torchvision/prototype/transforms/_misc.py

@@ -7,8 +7,8 @@ import torch
 from torchvision.ops import remove_small_boxes
 from torchvision.prototype import features
 from torchvision.prototype.transforms import functional as F, Transform
-from torchvision.prototype.transforms._utils import has_any, query_bounding_box
-from torchvision.transforms.transforms import _setup_size
+
+from ._utils import _setup_size, has_any, query_bounding_box
 
 
 class Identity(Transform):

torchvision/prototype/transforms/_type_conversion.py

@@ -52,3 +52,9 @@ class ToImagePIL(Transform):
 
     def _transform(self, inpt: Any, params: Dict[str, Any]) -> PIL.Image.Image:
         return F.to_image_pil(inpt, mode=self.mode)
+
+
+# We changed the names to align them with the new naming scheme. Still, `PILToTensor` and `ToPILImage` are
+# prevalent and well understood. Thus, we just alias them without deprecating the old names.
+PILToTensor = ToImageTensor
+ToPILImage = ToImagePIL

torchvision/prototype/transforms/_utils.py

 from typing import Any, Callable, Tuple, Type, Union
 
 import PIL.Image
-import torch
 from torch.utils._pytree import tree_flatten
 from torchvision._utils import sequence_to_str
 from torchvision.prototype import features
 
-from .functional._meta import get_dimensions_image_pil, get_dimensions_image_tensor
+from torchvision.prototype.transforms.functional._meta import get_chw
+from torchvision.transforms.functional_tensor import _parse_pad_padding  # noqa: F401
+from torchvision.transforms.transforms import _check_sequence_input, _setup_angle, _setup_size  # noqa: F401
 
 
 def query_bounding_box(sample: Any) -> features.BoundingBox:
@@ -19,19 +20,6 @@ def query_bounding_box(sample: Any) -> features.BoundingBox:
     return bounding_boxes.pop()
 
 
-def get_chw(image: Union[PIL.Image.Image, torch.Tensor, features.Image]) -> Tuple[int, int, int]:
-    if isinstance(image, features.Image):
-        channels = image.num_channels
-        height, width = image.image_size
-    elif features.is_simple_tensor(image):
-        channels, height, width = get_dimensions_image_tensor(image)
-    elif isinstance(image, PIL.Image.Image):
-        channels, height, width = get_dimensions_image_pil(image)
-    else:
-        raise TypeError(f"unable to get image dimensions from object of type {type(image).__name__}")
-    return channels, height, width
-
-
 def query_chw(sample: Any) -> Tuple[int, int, int]:
     flat_sample, _ = tree_flatten(sample)
     chws = {

torchvision/prototype/transforms/functional/__init__.py

@@ -5,6 +5,9 @@ from ._meta import (
     convert_color_space_image_tensor,
     convert_color_space_image_pil,
     convert_color_space,
+    get_dimensions,
+    get_image_num_channels,
+    get_image_size,
 )  # usort: skip
 
 from ._augment import erase, erase_image_pil, erase_image_tensor
@@ -68,6 +71,7 @@ from ._geometry import (
     five_crop,
     five_crop_image_pil,
     five_crop_image_tensor,
+    hflip,
     horizontal_flip,
     horizontal_flip_bounding_box,
     horizontal_flip_image_pil,
@@ -106,8 +110,17 @@ from ._geometry import (
     vertical_flip_image_pil,
     vertical_flip_image_tensor,
     vertical_flip_segmentation_mask,
+    vflip,
 )
 from ._misc import gaussian_blur, gaussian_blur_image_pil, gaussian_blur_image_tensor, normalize, normalize_image_tensor
-from ._type_conversion import decode_image_with_pil, decode_video_with_av, to_image_pil, to_image_tensor
+from ._type_conversion import (
+    convert_image_dtype,
+    decode_image_with_pil,
+    decode_video_with_av,
+    pil_to_tensor,
+    to_image_pil,
+    to_image_tensor,
+    to_pil_image,
+)
 
 from ._deprecated import rgb_to_grayscale, to_grayscale  # usort: skip

torchvision/prototype/transforms/functional/_deprecated.py

@@ -2,6 +2,7 @@ import warnings
 from typing import Any
 
 import PIL.Image
+import torch
 
 from torchvision.prototype import features
 from torchvision.transforms import functional as _F
@@ -39,3 +40,11 @@ def rgb_to_grayscale(inpt: Any, num_output_channels: int = 1) -> Any:
     )
 
     return _F.rgb_to_grayscale(inpt, num_output_channels=num_output_channels)
+
+
+def to_tensor(inpt: Any) -> torch.Tensor:
+    warnings.warn(
+        "The function `to_tensor(...)` is deprecated and will be removed in a future release. "
+        "Instead, please use `to_image_tensor(...)` followed by `convert_image_dtype(...)`."
+    )
+    return _F.to_tensor(inpt)

torchvision/prototype/transforms/functional/_geometry.py

@@ -89,6 +89,12 @@ def vertical_flip(inpt: DType) -> DType:
         return vertical_flip_image_tensor(inpt)
 
 
+# We changed the names to align them with the transforms, i.e. `RandomHorizontalFlip`. Still, `hflip` and `vflip` are
+# prevalent and well understood. Thus, we just alias them without deprecating the old names.
+hflip = horizontal_flip
+vflip = vertical_flip
+
+
 def resize_image_tensor(
     image: torch.Tensor,
     size: List[int],

torchvision/prototype/transforms/functional/_meta.py

-from typing import Any, Optional, Tuple
+from typing import Any, List, Optional, Tuple, Union
 
 import PIL.Image
 import torch
+from torchvision.prototype import features
 from torchvision.prototype.features import BoundingBoxFormat, ColorSpace, Image
 from torchvision.transforms import functional_pil as _FP, functional_tensor as _FT
 
@@ -9,6 +10,40 @@ get_dimensions_image_tensor = _FT.get_dimensions
 get_dimensions_image_pil = _FP.get_dimensions
 
 
+def get_chw(image: Union[PIL.Image.Image, torch.Tensor, features.Image]) -> Tuple[int, int, int]:
+    if isinstance(image, features.Image):
+        channels = image.num_channels
+        height, width = image.image_size
+    elif features.is_simple_tensor(image):
+        channels, height, width = get_dimensions_image_tensor(image)
+    elif isinstance(image, PIL.Image.Image):
+        channels, height, width = get_dimensions_image_pil(image)
+    else:
+        raise TypeError(f"unable to get image dimensions from object of type {type(image).__name__}")
+    return channels, height, width
+
+
+# The three functions below are here for BC. Whether we want to have two different kernels and how they and the
+# compound version should be named is still under discussion: https://github.com/pytorch/vision/issues/6491
+# Given that these kernels should also support boxes, masks, and videos, it is unlikely that there name will stay.
+# They will either be deprecated or simply aliased to the new kernels if we have reached consensus about the issue
+# detailed above.
+
+
+def get_dimensions(image: Union[PIL.Image.Image, torch.Tensor, features.Image]) -> List[int]:
+    return list(get_chw(image))
+
+
+def get_image_num_channels(image: Union[PIL.Image.Image, torch.Tensor, features.Image]) -> int:
+    num_channels, *_ = get_chw(image)
+    return num_channels
+
+
+def get_image_size(image: Union[PIL.Image.Image, torch.Tensor, features.Image]) -> List[int]:
+    _, *image_size = get_chw(image)
+    return image_size
+
+
 def _xywh_to_xyxy(xywh: torch.Tensor) -> torch.Tensor:
     xyxy = xywh.clone()
     xyxy[..., 2:] += xyxy[..., :2]

torchvision/prototype/transforms/functional/_type_conversion.py

@@ -31,3 +31,10 @@ def to_image_tensor(image: Union[torch.Tensor, PIL.Image.Image, np.ndarray]) ->
 
 
 to_image_pil = _F.to_pil_image
+
+# We changed the names to align them with the new naming scheme. Still, `to_pil_image` and `pil_to_tensor` are
+# prevalent and well understood. Thus, we just alias them without deprecating the old names.
+to_pil_image = to_image_pil
+pil_to_tensor = to_image_tensor
+
+convert_image_dtype = _F.convert_image_dtype