Pad CV-CUDA Backend

test/test_transforms_v2.py

@@ -21,6 +21,7 @@
 import torchvision.transforms.v2 as transforms
 
 from common_utils import (
+    assert_close,
     assert_equal,
     cache,
     cpu_and_cuda,
@@ -41,7 +42,6 @@
 )
 
 from torch import nn
-from torch.testing import assert_close
 from torch.utils._pytree import tree_flatten, tree_map
 from torch.utils.data import DataLoader, default_collate
 from torchvision import tv_tensors
@@ -4727,6 +4727,9 @@ def test_kernel_video(self):
             make_segmentation_mask,
             make_video,
             make_keypoints,
+            pytest.param(
+                make_image_cvcuda, marks=pytest.mark.skipif(not CVCUDA_AVAILABLE, reason="test requires CVCUDA")
+            ),
         ],
     )
     def test_functional(self, make_input):
@@ -4745,9 +4748,16 @@ def test_functional(self, make_input):
             (F.pad_bounding_boxes, tv_tensors.BoundingBoxes),
             (F.pad_mask, tv_tensors.Mask),
             (F.pad_video, tv_tensors.Video),
+            pytest.param(
+                F._geometry._pad_image_cvcuda,
+                None,
+                marks=pytest.mark.skipif(not CVCUDA_AVAILABLE, reason="test requires CVCUDA"),
+            ),
         ],
     )
     def test_functional_signature(self, kernel, input_type):
+        if kernel is F._geometry._pad_image_cvcuda:
+            input_type = _import_cvcuda().Tensor
         check_functional_kernel_signature_match(F.pad, kernel=kernel, input_type=input_type)
 
     @pytest.mark.parametrize(
@@ -4760,6 +4770,9 @@ def test_functional_signature(self, kernel, input_type):
             make_segmentation_mask,
             make_video,
             make_keypoints,
+            pytest.param(
+                make_image_cvcuda, marks=pytest.mark.skipif(not CVCUDA_AVAILABLE, reason="test requires CVCUDA")
+            ),
         ],
     )
     def test_transform(self, make_input):
@@ -4784,6 +4797,15 @@ def test_transform_errors(self):
         with pytest.raises(ValueError, match="Padding mode should be either"):
             transforms.Pad(12, padding_mode="abc")
 
+    @pytest.mark.parametrize(
+        "make_input",
+        [
+            make_image,
+            pytest.param(
+                make_image_cvcuda, marks=pytest.mark.skipif(not CVCUDA_AVAILABLE, reason="test requires CVCUDA")
+            ),
+        ],
+    )
     @pytest.mark.parametrize("padding", CORRECTNESS_PADDINGS)
     @pytest.mark.parametrize(
         ("padding_mode", "fill"),
@@ -4793,12 +4815,16 @@ def test_transform_errors(self):
         ],
     )
     @pytest.mark.parametrize("fn", [F.pad, transform_cls_to_functional(transforms.Pad)])
-    def test_image_correctness(self, padding, padding_mode, fill, fn):
-        image = make_image(dtype=torch.uint8, device="cpu")
+    def test_image_correctness(self, make_input, padding, padding_mode, fill, fn):
+        image = make_input(dtype=torch.uint8, device="cpu")
 
         fill = adapt_fill(fill, dtype=torch.uint8)
 
         actual = fn(image, padding=padding, padding_mode=padding_mode, fill=fill)
+
+        if make_input is make_image_cvcuda:
+            image = F.cvcuda_to_tensor(image)[0].cpu()
+
         expected = F.to_image(F.pad(F.to_pil_image(image), padding=padding, padding_mode=padding_mode, fill=fill))
 
         assert_equal(actual, expected)

torchvision/transforms/v2/_geometry.py

@@ -465,6 +465,9 @@ class Pad(Transform):
 
     _v1_transform_cls = _transforms.Pad
 
+    if CVCUDA_AVAILABLE:
+        _transformed_types = Transform._transformed_types + (_is_cvcuda_tensor,)
+
     def _extract_params_for_v1_transform(self) -> dict[str, Any]:
         params = super()._extract_params_for_v1_transform()
 

torchvision/transforms/v2/_utils.py

@@ -16,7 +16,7 @@
 
 from torchvision.transforms.transforms import _check_sequence_input, _setup_angle, _setup_size  # noqa: F401
 from torchvision.transforms.v2.functional import get_dimensions, get_size, is_pure_tensor
-from torchvision.transforms.v2.functional._utils import _FillType, _FillTypeJIT
+from torchvision.transforms.v2.functional._utils import _FillType, _FillTypeJIT, _is_cvcuda_tensor
 
 
 def _setup_number_or_seq(arg: int | float | Sequence[int | float], name: str) -> Sequence[float]:
@@ -182,7 +182,7 @@ def query_chw(flat_inputs: list[Any]) -> tuple[int, int, int]:
     chws = {
         tuple(get_dimensions(inpt))
         for inpt in flat_inputs
-        if check_type(inpt, (is_pure_tensor, tv_tensors.Image, PIL.Image.Image, tv_tensors.Video))
+        if check_type(inpt, (is_pure_tensor, tv_tensors.Image, PIL.Image.Image, tv_tensors.Video, _is_cvcuda_tensor))
     }
     if not chws:
         raise TypeError("No image or video was found in the sample")
@@ -207,6 +207,7 @@ def query_size(flat_inputs: list[Any]) -> tuple[int, int]:
                 tv_tensors.Mask,
                 tv_tensors.BoundingBoxes,
                 tv_tensors.KeyPoints,
+                _is_cvcuda_tensor,
             ),
         )
     }

torchvision/transforms/v2/functional/_geometry.py

@@ -28,6 +28,7 @@
 
 from ._utils import (
     _FillTypeJIT,
+    _get_cvcuda_border_from_pad_mode,
     _get_kernel,
     _import_cvcuda,
     _is_cvcuda_available,
@@ -1682,6 +1683,38 @@ def _pad_with_vector_fill(
 _pad_image_pil = _register_kernel_internal(pad, PIL.Image.Image)(_FP.pad)
 
 
+def _pad_image_cvcuda(
+    image: "cvcuda.Tensor",
+    padding: list[int],
+    fill: Optional[Union[int, float, list[float]]] = None,
+    padding_mode: str = "constant",
+) -> "cvcuda.Tensor":
+    cvcuda = _import_cvcuda()
+
+    border_mode = _get_cvcuda_border_from_pad_mode(padding_mode)
+
+    if fill is None:
+        fill = 0
+    if isinstance(fill, (int, float)):
+        fill = [fill] * image.shape[3]
+
+    left, right, top, bottom = _parse_pad_padding(padding)
+
+    return cvcuda.copymakeborder(
+        image,
+        border_mode=border_mode,
+        border_value=fill,
+        top=top,
+        left=left,
+        bottom=bottom,
+        right=right,
+    )
+
+
+if CVCUDA_AVAILABLE:
+    _register_kernel_internal(pad, _import_cvcuda().Tensor)(_pad_image_cvcuda)
+
+
 @_register_kernel_internal(pad, tv_tensors.Mask)
 def pad_mask(
     mask: torch.Tensor,

torchvision/transforms/v2/functional/_meta.py

@@ -51,6 +51,16 @@ def get_dimensions_video(video: torch.Tensor) -> list[int]:
     return get_dimensions_image(video)
 
 
+def get_dimensions_image_cvcuda(image: "cvcuda.Tensor") -> list[int]:
+    # CV-CUDA tensor is always in NHWC layout
+    # get_dimensions is CHW
+    return [image.shape[3], image.shape[1], image.shape[2]]
+
+
+if CVCUDA_AVAILABLE:
+    _register_kernel_internal(get_dimensions, cvcuda.Tensor)(get_dimensions_image_cvcuda)
+
+
 def get_num_channels(inpt: torch.Tensor) -> int:
     if torch.jit.is_scripting():
         return get_num_channels_image(inpt)
@@ -87,6 +97,16 @@ def get_num_channels_video(video: torch.Tensor) -> int:
 get_image_num_channels = get_num_channels
 
 
+def get_num_channels_image_cvcuda(image: "cvcuda.Tensor") -> int:
+    # CV-CUDA tensor is always in NHWC layout
+    # get_num_channels is C
+    return image.shape[3]
+
+
+if CVCUDA_AVAILABLE:
+    _register_kernel_internal(get_num_channels, cvcuda.Tensor)(get_num_channels_image_cvcuda)
+
+
 def get_size(inpt: torch.Tensor) -> list[int]:
     if torch.jit.is_scripting():
         return get_size_image(inpt)
@@ -125,7 +145,7 @@ def get_size_image_cvcuda(image: "cvcuda.Tensor") -> list[int]:
 
 
 if CVCUDA_AVAILABLE:
-    _get_size_image_cvcuda = _register_kernel_internal(get_size, cvcuda.Tensor)(get_size_image_cvcuda)
+    _register_kernel_internal(get_size, _import_cvcuda().Tensor)(get_size_image_cvcuda)
 
 
 @_register_kernel_internal(get_size, tv_tensors.Video, tv_tensor_wrapper=False)

torchvision/transforms/v2/functional/_misc.py

@@ -1,5 +1,5 @@
 import math
-from typing import Optional
+from typing import Optional, TYPE_CHECKING
 
 import PIL.Image
 import torch
@@ -13,7 +13,14 @@
 
 from ._meta import _convert_bounding_box_format
 
-from ._utils import _get_kernel, _register_kernel_internal, is_pure_tensor
+from ._utils import _get_kernel, _import_cvcuda, _is_cvcuda_available, _register_kernel_internal, is_pure_tensor
+
+CVCUDA_AVAILABLE = _is_cvcuda_available()
+
+if TYPE_CHECKING:
+    import cvcuda  # type: ignore[import-not-found]
+if CVCUDA_AVAILABLE:
+    cvcuda = _import_cvcuda()  # noqa: F811
 
 
 def normalize(

torchvision/transforms/v2/functional/_utils.py

@@ -1,10 +1,13 @@
 import functools
 from collections.abc import Sequence
-from typing import Any, Callable, Optional, Union
+from typing import Any, Callable, Optional, TYPE_CHECKING, Union
 
 import torch
 from torchvision import tv_tensors
 
+if TYPE_CHECKING:
+    import cvcuda  # type: ignore[import-not-found]
+
 _FillType = Union[int, float, Sequence[int], Sequence[float], None]
 _FillTypeJIT = Optional[list[float]]
 
@@ -177,3 +180,22 @@ def _is_cvcuda_tensor(inpt: Any) -> bool:
         return isinstance(inpt, cvcuda.Tensor)
     except ImportError:
         return False
+
+
+_pad_mode_to_cvcuda_border: dict[str, "cvcuda.Border"] = {}
+
+
+def _get_cvcuda_border_from_pad_mode(pad_mode: str) -> "cvcuda.Border":
+    if len(_pad_mode_to_cvcuda_border) == 0:
+        cvcuda = _import_cvcuda()
+        _pad_mode_to_cvcuda_border["constant"] = cvcuda.Border.CONSTANT
+        _pad_mode_to_cvcuda_border["reflect"] = cvcuda.Border.REFLECT101
+        _pad_mode_to_cvcuda_border["replicate"] = cvcuda.Border.REPLICATE
+        _pad_mode_to_cvcuda_border["edge"] = cvcuda.Border.REPLICATE
+        _pad_mode_to_cvcuda_border["symmetric"] = cvcuda.Border.REFLECT
+
+    border_mode = _pad_mode_to_cvcuda_border.get(pad_mode)
+    if border_mode is None:
+        raise ValueError(f"Pad mode {pad_mode} is not supported with CV-CUDA")
+
+    return border_mode