Update transforms for PIL deprecation (#5898)

* Update transforms for PIL deprecation

* Changes agreed at pytorch/vision#5898

* black, sort constants, version check

* Format tests

* Square brackets

* Update torchvision/transforms/_pil_constants.py
Co-authored-by: Nicolas Hug <contact@nicolas-hug.com>
Co-authored-by: Philip Meier <github.pmeier@posteo.de>
Co-authored-by: Nicolas Hug <contact@nicolas-hug.com>
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

test/test_onnx.py

@@ -412,12 +412,13 @@ class TestONNXExporter:
     def get_image(self, rel_path: str, size: Tuple[int, int]) -> torch.Tensor:
         import os
 
+        import torchvision.transforms._pil_constants as _pil_constants
         from PIL import Image
         from torchvision.transforms import functional as F
 
         data_dir = os.path.join(os.path.dirname(__file__), "assets")
         path = os.path.join(data_dir, *rel_path.split("/"))
-        image = Image.open(path).convert("RGB").resize(size, Image.BILINEAR)
+        image = Image.open(path).convert("RGB").resize(size, _pil_constants.BILINEAR)
 
         return F.convert_image_dtype(F.pil_to_tensor(image))
 

test/test_transforms.py

@@ -8,6 +8,7 @@ import numpy as np
 import pytest
 import torch
 import torchvision.transforms as transforms
+import torchvision.transforms._pil_constants as _pil_constants
 import torchvision.transforms.functional as F
 import torchvision.transforms.functional_tensor as F_t
 from PIL import Image
@@ -173,7 +174,7 @@ class TestAccImage:
     def test_accimage_resize(self):
         trans = transforms.Compose(
             [
-                transforms.Resize(256, interpolation=Image.LINEAR),
+                transforms.Resize(256, interpolation=_pil_constants.LINEAR),
                 transforms.PILToTensor(),
                 transforms.ConvertImageDtype(dtype=torch.float),
             ]

test/test_transforms_tensor.py

@@ -4,6 +4,7 @@ import sys
 import numpy as np
 import pytest
 import torch
+import torchvision.transforms._pil_constants as _pil_constants
 from common_utils import (
     get_tmp_dir,
     int_dtypes,
@@ -15,7 +16,6 @@ from common_utils import (
     cpu_and_gpu,
     assert_equal,
 )
-from PIL import Image
 from torchvision import transforms as T
 from torchvision.transforms import InterpolationMode
 from torchvision.transforms import functional as F
@@ -771,13 +771,13 @@ def test_autoaugment__op_apply_shear(interpolation, mode):
             matrix = (1, level, 0, 0, 1, 0)
         elif mode == "Y":
             matrix = (1, 0, 0, level, 1, 0)
-        return pil_img.transform((image_size, image_size), Image.AFFINE, matrix, resample=resample)
+        return pil_img.transform((image_size, image_size), _pil_constants.AFFINE, matrix, resample=resample)
 
     t_img, pil_img = _create_data(image_size, image_size)
 
     resample_pil = {
-        F.InterpolationMode.NEAREST: Image.NEAREST,
-        F.InterpolationMode.BILINEAR: Image.BILINEAR,
+        F.InterpolationMode.NEAREST: _pil_constants.NEAREST,
+        F.InterpolationMode.BILINEAR: _pil_constants.BILINEAR,
     }[interpolation]
 
     level = 0.3

torchvision/transforms/_pil_constants.py

+import PIL
+from PIL import Image
+
+# See https://pillow.readthedocs.io/en/stable/releasenotes/9.1.0.html#deprecations
+# TODO: Remove this file once PIL minimal version is >= 9.1
+
+if tuple(int(part) for part in PIL.__version__.split(".")) >= (9, 1):
+    BICUBIC = Image.Resampling.BICUBIC
+    BILINEAR = Image.Resampling.BILINEAR
+    LINEAR = Image.Resampling.BILINEAR
+    NEAREST = Image.Resampling.NEAREST
+
+    AFFINE = Image.Transform.AFFINE
+    FLIP_LEFT_RIGHT = Image.Transpose.FLIP_LEFT_RIGHT
+    FLIP_TOP_BOTTOM = Image.Transpose.FLIP_TOP_BOTTOM
+    PERSPECTIVE = Image.Transform.PERSPECTIVE
+else:
+    BICUBIC = Image.BICUBIC
+    BILINEAR = Image.BILINEAR
+    NEAREST = Image.NEAREST
+    LINEAR = Image.LINEAR
+
+    AFFINE = Image.AFFINE
+    FLIP_LEFT_RIGHT = Image.FLIP_LEFT_RIGHT
+    FLIP_TOP_BOTTOM = Image.FLIP_TOP_BOTTOM
+    PERSPECTIVE = Image.PERSPECTIVE

torchvision/transforms/functional.py

@@ -392,7 +392,7 @@ def resize(
             :class:`torchvision.transforms.InterpolationMode`.
             Default is ``InterpolationMode.BILINEAR``. If input is Tensor, only ``InterpolationMode.NEAREST``,
             ``InterpolationMode.BILINEAR`` and ``InterpolationMode.BICUBIC`` are supported.
-            For backward compatibility integer values (e.g. ``PIL.Image.NEAREST``) are still acceptable.
+            For backward compatibility integer values (e.g. ``PIL.Image[.Resampling].NEAREST``) are still acceptable.
         max_size (int, optional): The maximum allowed for the longer edge of
             the resized image: if the longer edge of the image is greater
             than ``max_size`` after being resized according to ``size``, then
@@ -572,7 +572,7 @@ def resized_crop(
             :class:`torchvision.transforms.InterpolationMode`.
             Default is ``InterpolationMode.BILINEAR``. If input is Tensor, only ``InterpolationMode.NEAREST``,
             ``InterpolationMode.BILINEAR`` and ``InterpolationMode.BICUBIC`` are supported.
-            For backward compatibility integer values (e.g. ``PIL.Image.NEAREST``) are still acceptable.
+            For backward compatibility integer values (e.g. ``PIL.Image[.Resampling].NEAREST``) are still acceptable.
 
     Returns:
         PIL Image or Tensor: Cropped image.
@@ -652,7 +652,7 @@ def perspective(
         interpolation (InterpolationMode): Desired interpolation enum defined by
             :class:`torchvision.transforms.InterpolationMode`. Default is ``InterpolationMode.BILINEAR``.
             If input is Tensor, only ``InterpolationMode.NEAREST``, ``InterpolationMode.BILINEAR`` are supported.
-            For backward compatibility integer values (e.g. ``PIL.Image.NEAREST``) are still acceptable.
+            For backward compatibility integer values (e.g. ``PIL.Image[.Resampling].NEAREST``) are still acceptable.
         fill (sequence or number, optional): Pixel fill value for the area outside the transformed
             image. If given a number, the value is used for all bands respectively.
 
@@ -1012,7 +1012,7 @@ def rotate(
         interpolation (InterpolationMode): Desired interpolation enum defined by
             :class:`torchvision.transforms.InterpolationMode`. Default is ``InterpolationMode.NEAREST``.
             If input is Tensor, only ``InterpolationMode.NEAREST``, ``InterpolationMode.BILINEAR`` are supported.
-            For backward compatibility integer values (e.g. ``PIL.Image.NEAREST``) are still acceptable.
+            For backward compatibility integer values (e.g. ``PIL.Image[.Resampling].NEAREST``) are still acceptable.
         expand (bool, optional): Optional expansion flag.
             If true, expands the output image to make it large enough to hold the entire rotated image.
             If false or omitted, make the output image the same size as the input image.
@@ -1105,7 +1105,7 @@ def affine(
         interpolation (InterpolationMode): Desired interpolation enum defined by
             :class:`torchvision.transforms.InterpolationMode`. Default is ``InterpolationMode.NEAREST``.
             If input is Tensor, only ``InterpolationMode.NEAREST``, ``InterpolationMode.BILINEAR`` are supported.
-            For backward compatibility integer values (e.g. ``PIL.Image.NEAREST``) are still acceptable.
+            For backward compatibility integer values (e.g. ``PIL.Image[.Resampling].NEAREST``) are still acceptable.
         fill (sequence or number, optional): Pixel fill value for the area outside the transformed
             image. If given a number, the value is used for all bands respectively.
 

torchvision/transforms/functional_pil.py

@@ -10,6 +10,7 @@ try:
     import accimage
 except ImportError:
     accimage = None
+from . import _pil_constants
 
 
 @torch.jit.unused
@@ -54,7 +55,7 @@ def hflip(img: Image.Image) -> Image.Image:
     if not _is_pil_image(img):
         raise TypeError(f"img should be PIL Image. Got {type(img)}")
 
-    return img.transpose(Image.FLIP_LEFT_RIGHT)
+    return img.transpose(_pil_constants.FLIP_LEFT_RIGHT)
 
 
 @torch.jit.unused
@@ -62,7 +63,7 @@ def vflip(img: Image.Image) -> Image.Image:
     if not _is_pil_image(img):
         raise TypeError(f"img should be PIL Image. Got {type(img)}")
 
-    return img.transpose(Image.FLIP_TOP_BOTTOM)
+    return img.transpose(_pil_constants.FLIP_TOP_BOTTOM)
 
 
 @torch.jit.unused
@@ -240,7 +241,7 @@ def crop(
 def resize(
     img: Image.Image,
     size: Union[Sequence[int], int],
-    interpolation: int = Image.BILINEAR,
+    interpolation: int = _pil_constants.BILINEAR,
     max_size: Optional[int] = None,
 ) -> Image.Image:
 
@@ -314,7 +315,7 @@ def _parse_fill(
 def affine(
     img: Image.Image,
     matrix: List[float],
-    interpolation: int = Image.NEAREST,
+    interpolation: int = _pil_constants.NEAREST,
     fill: Optional[Union[float, List[float], Tuple[float, ...]]] = 0,
 ) -> Image.Image:
 
@@ -323,14 +324,14 @@ def affine(
 
     output_size = img.size
     opts = _parse_fill(fill, img)
-    return img.transform(output_size, Image.AFFINE, matrix, interpolation, **opts)
+    return img.transform(output_size, _pil_constants.AFFINE, matrix, interpolation, **opts)
 
 
 @torch.jit.unused
 def rotate(
     img: Image.Image,
     angle: float,
-    interpolation: int = Image.NEAREST,
+    interpolation: int = _pil_constants.NEAREST,
     expand: bool = False,
     center: Optional[Tuple[int, int]] = None,
     fill: Optional[Union[float, List[float], Tuple[float, ...]]] = 0,
@@ -347,7 +348,7 @@ def rotate(
 def perspective(
     img: Image.Image,
     perspective_coeffs: float,
-    interpolation: int = Image.BICUBIC,
+    interpolation: int = _pil_constants.BICUBIC,
     fill: Optional[Union[float, List[float], Tuple[float, ...]]] = 0,
 ) -> Image.Image:
 
@@ -356,7 +357,7 @@ def perspective(
 
     opts = _parse_fill(fill, img)
 
-    return img.transform(img.size, Image.PERSPECTIVE, perspective_coeffs, interpolation, **opts)
+    return img.transform(img.size, _pil_constants.PERSPECTIVE, perspective_coeffs, interpolation, **opts)
 
 
 @torch.jit.unused

torchvision/transforms/transforms.py

@@ -17,7 +17,6 @@ from ..utils import _log_api_usage_once
 from . import functional as F
 from .functional import InterpolationMode, _interpolation_modes_from_int
 
-
 __all__ = [
     "Compose",
     "ToTensor",
@@ -298,7 +297,7 @@ class Resize(torch.nn.Module):
             :class:`torchvision.transforms.InterpolationMode`. Default is ``InterpolationMode.BILINEAR``.
             If input is Tensor, only ``InterpolationMode.NEAREST``, ``InterpolationMode.BILINEAR`` and
             ``InterpolationMode.BICUBIC`` are supported.
-            For backward compatibility integer values (e.g. ``PIL.Image.NEAREST``) are still acceptable.
+            For backward compatibility integer values (e.g. ``PIL.Image[.Resampling].NEAREST``) are still acceptable.
         max_size (int, optional): The maximum allowed for the longer edge of
             the resized image: if the longer edge of the image is greater
             than ``max_size`` after being resized according to ``size``, then
@@ -755,7 +754,7 @@ class RandomPerspective(torch.nn.Module):
         interpolation (InterpolationMode): Desired interpolation enum defined by
             :class:`torchvision.transforms.InterpolationMode`. Default is ``InterpolationMode.BILINEAR``.
             If input is Tensor, only ``InterpolationMode.NEAREST``, ``InterpolationMode.BILINEAR`` are supported.
-            For backward compatibility integer values (e.g. ``PIL.Image.NEAREST``) are still acceptable.
+            For backward compatibility integer values (e.g. ``PIL.Image[.Resampling].NEAREST``) are still acceptable.
         fill (sequence or number): Pixel fill value for the area outside the transformed
             image. Default is ``0``. If given a number, the value is used for all bands respectively.
     """
@@ -869,7 +868,7 @@ class RandomResizedCrop(torch.nn.Module):
             :class:`torchvision.transforms.InterpolationMode`. Default is ``InterpolationMode.BILINEAR``.
             If input is Tensor, only ``InterpolationMode.NEAREST``, ``InterpolationMode.BILINEAR`` and
             ``InterpolationMode.BICUBIC`` are supported.
-            For backward compatibility integer values (e.g. ``PIL.Image.NEAREST``) are still acceptable.
+            For backward compatibility integer values (e.g. ``PIL.Image[.Resampling].NEAREST``) are still acceptable.
 
     """
 
@@ -1268,7 +1267,7 @@ class RandomRotation(torch.nn.Module):
         interpolation (InterpolationMode): Desired interpolation enum defined by
             :class:`torchvision.transforms.InterpolationMode`. Default is ``InterpolationMode.NEAREST``.
             If input is Tensor, only ``InterpolationMode.NEAREST``, ``InterpolationMode.BILINEAR`` are supported.
-            For backward compatibility integer values (e.g. ``PIL.Image.NEAREST``) are still acceptable.
+            For backward compatibility integer values (e.g. ``PIL.Image[.Resampling].NEAREST``) are still acceptable.
         expand (bool, optional): Optional expansion flag.
             If true, expands the output to make it large enough to hold the entire rotated image.
             If false or omitted, make the output image the same size as the input image.
@@ -1389,7 +1388,7 @@ class RandomAffine(torch.nn.Module):
         interpolation (InterpolationMode): Desired interpolation enum defined by
             :class:`torchvision.transforms.InterpolationMode`. Default is ``InterpolationMode.NEAREST``.
             If input is Tensor, only ``InterpolationMode.NEAREST``, ``InterpolationMode.BILINEAR`` are supported.
-            For backward compatibility integer values (e.g. ``PIL.Image.NEAREST``) are still acceptable.
+            For backward compatibility integer values (e.g. ``PIL.Image[.Resampling].NEAREST``) are still acceptable.
         fill (sequence or number): Pixel fill value for the area outside the transformed
             image. Default is ``0``. If given a number, the value is used for all bands respectively.
         fillcolor (sequence or number, optional):