From 44db71c0772e5ef5758c38d0e4e8ad9995946c80 Mon Sep 17 00:00:00 2001
From: Justin Davis <judavis@nvidia.com>
Date: Tue, 25 Nov 2025 09:14:49 -0800
Subject: [PATCH 01/18] implement additional cvcuda infra for all branches to
 avoid duplicate setup

---
 torchvision/transforms/v2/_transform.py       |  4 ++--
 torchvision/transforms/v2/_utils.py           |  3 ++-
 .../transforms/v2/functional/__init__.py      |  2 +-
 .../transforms/v2/functional/_augment.py      | 11 ++++++++++-
 .../transforms/v2/functional/_color.py        | 12 +++++++++++-
 .../transforms/v2/functional/_geometry.py     | 19 +++++++++++++++++--
 torchvision/transforms/v2/functional/_misc.py | 11 +++++++++--
 .../transforms/v2/functional/_utils.py        | 16 ++++++++++++++++
 8 files changed, 68 insertions(+), 10 deletions(-)

diff --git a/torchvision/transforms/v2/_transform.py b/torchvision/transforms/v2/_transform.py
index ac84fcb6c82..bec9ffcf714 100644
--- a/torchvision/transforms/v2/_transform.py
+++ b/torchvision/transforms/v2/_transform.py
@@ -11,7 +11,7 @@
 from torchvision.transforms.v2._utils import check_type, has_any, is_pure_tensor
 from torchvision.utils import _log_api_usage_once
 
-from .functional._utils import _get_kernel
+from .functional._utils import _get_kernel, is_cvcuda_tensor
 
 
 class Transform(nn.Module):
@@ -23,7 +23,7 @@ class Transform(nn.Module):
 
     # Class attribute defining transformed types. Other types are passed-through without any transformation
     # We support both Types and callables that are able to do further checks on the type of the input.
-    _transformed_types: tuple[type | Callable[[Any], bool], ...] = (torch.Tensor, PIL.Image.Image)
+    _transformed_types: tuple[type | Callable[[Any], bool], ...] = (torch.Tensor, PIL.Image.Image, is_cvcuda_tensor)
 
     def __init__(self) -> None:
         super().__init__()
diff --git a/torchvision/transforms/v2/_utils.py b/torchvision/transforms/v2/_utils.py
index bb6051b4e61..765a772fe41 100644
--- a/torchvision/transforms/v2/_utils.py
+++ b/torchvision/transforms/v2/_utils.py
@@ -15,7 +15,7 @@
 from torchvision._utils import sequence_to_str
 
 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 import get_dimensions, get_size, is_cvcuda_tensor, is_pure_tensor
 from torchvision.transforms.v2.functional._utils import _FillType, _FillTypeJIT
 
 
@@ -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,
             ),
         )
     }
diff --git a/torchvision/transforms/v2/functional/__init__.py b/torchvision/transforms/v2/functional/__init__.py
index 032a993b1f0..52181e4624b 100644
--- a/torchvision/transforms/v2/functional/__init__.py
+++ b/torchvision/transforms/v2/functional/__init__.py
@@ -1,6 +1,6 @@
 from torchvision.transforms import InterpolationMode  # usort: skip
 
-from ._utils import is_pure_tensor, register_kernel  # usort: skip
+from ._utils import is_pure_tensor, register_kernel, is_cvcuda_tensor  # usort: skip
 
 from ._meta import (
     clamp_bounding_boxes,
diff --git a/torchvision/transforms/v2/functional/_augment.py b/torchvision/transforms/v2/functional/_augment.py
index a904d8d7cbd..7ce5bdc7b7e 100644
--- a/torchvision/transforms/v2/functional/_augment.py
+++ b/torchvision/transforms/v2/functional/_augment.py
@@ -1,4 +1,5 @@
 import io
+from typing import TYPE_CHECKING
 
 import PIL.Image
 
@@ -8,7 +9,15 @@
 from torchvision.transforms.functional import pil_to_tensor, to_pil_image
 from torchvision.utils import _log_api_usage_once
 
-from ._utils import _get_kernel, _register_kernel_internal
+from ._utils import _get_kernel, _import_cvcuda, _is_cvcuda_available, _register_kernel_internal
+
+
+CVCUDA_AVAILABLE = _is_cvcuda_available()
+
+if TYPE_CHECKING:
+    import cvcuda  # type: ignore[import-not-found]
+if CVCUDA_AVAILABLE:
+    cvcuda = _import_cvcuda()  # noqa: F811
 
 
 def erase(
diff --git a/torchvision/transforms/v2/functional/_color.py b/torchvision/transforms/v2/functional/_color.py
index be254c0d63a..5be9c62902a 100644
--- a/torchvision/transforms/v2/functional/_color.py
+++ b/torchvision/transforms/v2/functional/_color.py
@@ -1,3 +1,5 @@
+from typing import TYPE_CHECKING
+
 import PIL.Image
 import torch
 from torch.nn.functional import conv2d
@@ -9,7 +11,15 @@
 
 from ._misc import _num_value_bits, to_dtype_image
 from ._type_conversion import pil_to_tensor, to_pil_image
-from ._utils import _get_kernel, _register_kernel_internal
+from ._utils import _get_kernel, _import_cvcuda, _is_cvcuda_available, _register_kernel_internal
+
+
+CVCUDA_AVAILABLE = _is_cvcuda_available()
+
+if TYPE_CHECKING:
+    import cvcuda  # type: ignore[import-not-found]
+if CVCUDA_AVAILABLE:
+    cvcuda = _import_cvcuda()  # noqa: F811
 
 
 def rgb_to_grayscale(inpt: torch.Tensor, num_output_channels: int = 1) -> torch.Tensor:
diff --git a/torchvision/transforms/v2/functional/_geometry.py b/torchvision/transforms/v2/functional/_geometry.py
index 4fcb7fabe0d..c029488001c 100644
--- a/torchvision/transforms/v2/functional/_geometry.py
+++ b/torchvision/transforms/v2/functional/_geometry.py
@@ -2,7 +2,7 @@
 import numbers
 import warnings
 from collections.abc import Sequence
-from typing import Any, Optional, Union
+from typing import Any, Optional, TYPE_CHECKING, Union
 
 import PIL.Image
 import torch
@@ -26,7 +26,22 @@
 
 from ._meta import _get_size_image_pil, clamp_bounding_boxes, convert_bounding_box_format
 
-from ._utils import _FillTypeJIT, _get_kernel, _register_five_ten_crop_kernel_internal, _register_kernel_internal
+from ._utils import (
+    _FillTypeJIT,
+    _get_kernel,
+    _import_cvcuda,
+    _is_cvcuda_available,
+    _register_five_ten_crop_kernel_internal,
+    _register_kernel_internal,
+)
+
+
+CVCUDA_AVAILABLE = _is_cvcuda_available()
+
+if TYPE_CHECKING:
+    import cvcuda  # type: ignore[import-not-found]
+if CVCUDA_AVAILABLE:
+    cvcuda = _import_cvcuda()  # noqa: F811
 
 
 def _check_interpolation(interpolation: Union[InterpolationMode, int]) -> InterpolationMode:
diff --git a/torchvision/transforms/v2/functional/_misc.py b/torchvision/transforms/v2/functional/_misc.py
index daf263df046..0fa05a2113c 100644
--- a/torchvision/transforms/v2/functional/_misc.py
+++ b/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(
diff --git a/torchvision/transforms/v2/functional/_utils.py b/torchvision/transforms/v2/functional/_utils.py
index ad1eddd258b..73fafaf7425 100644
--- a/torchvision/transforms/v2/functional/_utils.py
+++ b/torchvision/transforms/v2/functional/_utils.py
@@ -169,3 +169,19 @@ def _is_cvcuda_available():
         return True
     except ImportError:
         return False
+
+
+def is_cvcuda_tensor(inpt: Any) -> bool:
+    """
+    Check if the input is a CVCUDA tensor.
+
+    Args:
+        inpt: The input to check.
+
+    Returns:
+        True if the input is a CV-CUDA tensor, False otherwise.
+    """
+    if _is_cvcuda_available():
+        cvcuda = _import_cvcuda()
+        return isinstance(inpt, cvcuda.Tensor)
+    return False

From e3dd70022fa1c87aca7a9a98068b6e13e802a375 Mon Sep 17 00:00:00 2001
From: Justin Davis <judavis@nvidia.com>
Date: Tue, 25 Nov 2025 09:26:19 -0800
Subject: [PATCH 02/18] update make_image_cvcuda to have default batch dim

---
 test/common_utils.py | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/test/common_utils.py b/test/common_utils.py
index 8c3c9dd58a8..e7bae60c41b 100644
--- a/test/common_utils.py
+++ b/test/common_utils.py
@@ -400,8 +400,9 @@ def make_image_pil(*args, **kwargs):
     return to_pil_image(make_image(*args, **kwargs))
 
 
-def make_image_cvcuda(*args, **kwargs):
-    return to_cvcuda_tensor(make_image(*args, **kwargs))
+def make_image_cvcuda(*args, batch_dims=(1,), **kwargs):
+    # explicitly default batch_dims to (1,) since to_cvcuda_tensor requires a batch dimension (ndims == 4)
+    return to_cvcuda_tensor(make_image(*args, batch_dims=batch_dims, **kwargs))
 
 
 def make_keypoints(canvas_size=DEFAULT_SIZE, *, num_points=4, dtype=None, device="cpu"):

From c035df1c6eaebcad25604f8c298a7d9eaf86864b Mon Sep 17 00:00:00 2001
From: Justin Davis <judavis@nvidia.com>
Date: Mon, 1 Dec 2025 18:16:27 -0800
Subject: [PATCH 03/18] add stanardized setup to main for easier updating of
 PRs and branches

---
 test/common_utils.py                          | 21 ++++++++++++++--
 test/test_transforms_v2.py                    |  2 +-
 torchvision/transforms/v2/_utils.py           |  2 +-
 torchvision/transforms/v2/functional/_meta.py | 24 +++++++++++++++++--
 4 files changed, 43 insertions(+), 6 deletions(-)

diff --git a/test/common_utils.py b/test/common_utils.py
index e7bae60c41b..3b889e93d2e 100644
--- a/test/common_utils.py
+++ b/test/common_utils.py
@@ -20,13 +20,15 @@
 from torch.testing._comparison import BooleanPair, NonePair, not_close_error_metas, NumberPair, TensorLikePair
 from torchvision import io, tv_tensors
 from torchvision.transforms._functional_tensor import _max_value as get_max_value
-from torchvision.transforms.v2.functional import to_cvcuda_tensor, to_image, to_pil_image
+from torchvision.transforms.v2.functional import cvcuda_to_tensor, to_cvcuda_tensor, to_image, to_pil_image
+from torchvision.transforms.v2.functional._utils import _import_cvcuda, _is_cvcuda_available
 from torchvision.utils import _Image_fromarray
 
 
 IN_OSS_CI = any(os.getenv(var) == "true" for var in ["CIRCLECI", "GITHUB_ACTIONS"])
 IN_RE_WORKER = os.environ.get("INSIDE_RE_WORKER") is not None
 IN_FBCODE = os.environ.get("IN_FBCODE_TORCHVISION") == "1"
+CVCUDA_AVAILABLE = _is_cvcuda_available()
 CUDA_NOT_AVAILABLE_MSG = "CUDA device not available"
 MPS_NOT_AVAILABLE_MSG = "MPS device not available"
 OSS_CI_GPU_NO_CUDA_MSG = "We're in an OSS GPU machine, and this test doesn't need cuda."
@@ -275,6 +277,17 @@ def combinations_grid(**kwargs):
     return [dict(zip(kwargs.keys(), values)) for values in itertools.product(*kwargs.values())]
 
 
+def cvcuda_to_pil_compatible_tensor(tensor: "cvcuda.Tensor") -> torch.Tensor:
+    tensor = cvcuda_to_tensor(tensor)
+    if tensor.ndim != 4:
+        raise ValueError(f"CV-CUDA Tensor should be 4 dimensional. Got {tensor.ndim} dimensions.")
+    if tensor.shape[0] != 1:
+        raise ValueError(
+            f"CV-CUDA Tensor should have batch dimension 1 for comparison with PIL.Image.Image. Got {tensor.shape[0]}."
+        )
+    return tensor.squeeze(0).cpu()
+
+
 class ImagePair(TensorLikePair):
     def __init__(
         self,
@@ -287,6 +300,11 @@ def __init__(
         if all(isinstance(input, PIL.Image.Image) for input in [actual, expected]):
             actual, expected = (to_image(input) for input in [actual, expected])
 
+        # handle check for CV-CUDA Tensors
+        if CVCUDA_AVAILABLE and isinstance(actual, _import_cvcuda().Tensor):
+            # Use the PIL compatible tensor, so we can always compare with PIL.Image.Image
+            actual = cvcuda_to_pil_compatible_tensor(actual)
+
         super().__init__(actual, expected, **other_parameters)
         self.mae = mae
 
@@ -401,7 +419,6 @@ def make_image_pil(*args, **kwargs):
 
 
 def make_image_cvcuda(*args, batch_dims=(1,), **kwargs):
-    # explicitly default batch_dims to (1,) since to_cvcuda_tensor requires a batch dimension (ndims == 4)
     return to_cvcuda_tensor(make_image(*args, batch_dims=batch_dims, **kwargs))
 
 
diff --git a/test/test_transforms_v2.py b/test/test_transforms_v2.py
index 670a9d00ffb..7eba65550da 100644
--- a/test/test_transforms_v2.py
+++ b/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
diff --git a/torchvision/transforms/v2/_utils.py b/torchvision/transforms/v2/_utils.py
index 765a772fe41..3fc33ce5964 100644
--- a/torchvision/transforms/v2/_utils.py
+++ b/torchvision/transforms/v2/_utils.py
@@ -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")
diff --git a/torchvision/transforms/v2/functional/_meta.py b/torchvision/transforms/v2/functional/_meta.py
index 6b8f19f12f4..ee562cb2aee 100644
--- a/torchvision/transforms/v2/functional/_meta.py
+++ b/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_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_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_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_cvcuda)
+
+
 def get_size(inpt: torch.Tensor) -> list[int]:
     if torch.jit.is_scripting():
         return get_size_image(inpt)
@@ -114,7 +134,7 @@ def _get_size_image_pil(image: PIL.Image.Image) -> list[int]:
     return [height, width]
 
 
-def get_size_image_cvcuda(image: "cvcuda.Tensor") -> list[int]:
+def _get_size_cvcuda(image: "cvcuda.Tensor") -> list[int]:
     """Get size of `cvcuda.Tensor` with NHWC layout."""
     hw = list(image.shape[-3:-1])
     ndims = len(hw)
@@ -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, cvcuda.Tensor)(_get_size_cvcuda)
 
 
 @_register_kernel_internal(get_size, tv_tensors.Video, tv_tensor_wrapper=False)

From 98d7dfb2059eaf2c10c3f549ea45f1d27875134c Mon Sep 17 00:00:00 2001
From: Justin Davis <judavis@nvidia.com>
Date: Mon, 1 Dec 2025 18:25:09 -0800
Subject: [PATCH 04/18] update is_cvcuda_tensor

---
 torchvision/transforms/v2/functional/_utils.py | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/torchvision/transforms/v2/functional/_utils.py b/torchvision/transforms/v2/functional/_utils.py
index 73fafaf7425..44b2edeaf2d 100644
--- a/torchvision/transforms/v2/functional/_utils.py
+++ b/torchvision/transforms/v2/functional/_utils.py
@@ -181,7 +181,8 @@ def is_cvcuda_tensor(inpt: Any) -> bool:
     Returns:
         True if the input is a CV-CUDA tensor, False otherwise.
     """
-    if _is_cvcuda_available():
+    try:
         cvcuda = _import_cvcuda()
         return isinstance(inpt, cvcuda.Tensor)
-    return False
+    except ImportError:
+        return False

From ddc116d13febdae1d53507bcde9f103a4c14eba7 Mon Sep 17 00:00:00 2001
From: Justin Davis <judavis@nvidia.com>
Date: Tue, 2 Dec 2025 12:37:03 -0800
Subject: [PATCH 05/18] add cvcuda to pil compatible to transforms by default

---
 test/test_transforms_v2.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/test/test_transforms_v2.py b/test/test_transforms_v2.py
index 7eba65550da..87166477669 100644
--- a/test/test_transforms_v2.py
+++ b/test/test_transforms_v2.py
@@ -25,6 +25,7 @@
     assert_equal,
     cache,
     cpu_and_cuda,
+    cvcuda_to_pil_compatible_tensor,
     freeze_rng_state,
     ignore_jit_no_profile_information_warning,
     make_bounding_boxes,

From e51dc7eabd254261347245f4492892fd0944aae5 Mon Sep 17 00:00:00 2001
From: Justin Davis <judavis@nvidia.com>
Date: Tue, 2 Dec 2025 12:46:23 -0800
Subject: [PATCH 06/18] remove cvcuda from transform class

---
 torchvision/transforms/v2/_transform.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/torchvision/transforms/v2/_transform.py b/torchvision/transforms/v2/_transform.py
index bec9ffcf714..ac84fcb6c82 100644
--- a/torchvision/transforms/v2/_transform.py
+++ b/torchvision/transforms/v2/_transform.py
@@ -11,7 +11,7 @@
 from torchvision.transforms.v2._utils import check_type, has_any, is_pure_tensor
 from torchvision.utils import _log_api_usage_once
 
-from .functional._utils import _get_kernel, is_cvcuda_tensor
+from .functional._utils import _get_kernel
 
 
 class Transform(nn.Module):
@@ -23,7 +23,7 @@ class Transform(nn.Module):
 
     # Class attribute defining transformed types. Other types are passed-through without any transformation
     # We support both Types and callables that are able to do further checks on the type of the input.
-    _transformed_types: tuple[type | Callable[[Any], bool], ...] = (torch.Tensor, PIL.Image.Image, is_cvcuda_tensor)
+    _transformed_types: tuple[type | Callable[[Any], bool], ...] = (torch.Tensor, PIL.Image.Image)
 
     def __init__(self) -> None:
         super().__init__()

From 4939355a2c7421eeba95d7f155fe7953066aec6d Mon Sep 17 00:00:00 2001
From: Justin Davis <judavis@nvidia.com>
Date: Thu, 4 Dec 2025 11:07:08 -0800
Subject: [PATCH 07/18] resolve more formatting naming

---
 torchvision/transforms/v2/functional/__init__.py | 2 +-
 torchvision/transforms/v2/functional/_meta.py    | 8 ++++----
 2 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/torchvision/transforms/v2/functional/__init__.py b/torchvision/transforms/v2/functional/__init__.py
index 52181e4624b..032a993b1f0 100644
--- a/torchvision/transforms/v2/functional/__init__.py
+++ b/torchvision/transforms/v2/functional/__init__.py
@@ -1,6 +1,6 @@
 from torchvision.transforms import InterpolationMode  # usort: skip
 
-from ._utils import is_pure_tensor, register_kernel, is_cvcuda_tensor  # usort: skip
+from ._utils import is_pure_tensor, register_kernel  # usort: skip
 
 from ._meta import (
     clamp_bounding_boxes,
diff --git a/torchvision/transforms/v2/functional/_meta.py b/torchvision/transforms/v2/functional/_meta.py
index e8630f788ca..af03ad018d4 100644
--- a/torchvision/transforms/v2/functional/_meta.py
+++ b/torchvision/transforms/v2/functional/_meta.py
@@ -51,14 +51,14 @@ def get_dimensions_video(video: torch.Tensor) -> list[int]:
     return get_dimensions_image(video)
 
 
-def _get_dimensions_cvcuda(image: "cvcuda.Tensor") -> list[int]:
+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_cvcuda)
+    _register_kernel_internal(get_dimensions, cvcuda.Tensor)(get_dimensions_image_cvcuda)
 
 
 def get_num_channels(inpt: torch.Tensor) -> int:
@@ -97,14 +97,14 @@ def get_num_channels_video(video: torch.Tensor) -> int:
 get_image_num_channels = get_num_channels
 
 
-def _get_num_channels_cvcuda(image: "cvcuda.Tensor") -> int:
+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_cvcuda)
+    _register_kernel_internal(get_num_channels, cvcuda.Tensor)(get_num_channels_image_cvcuda)
 
 
 def get_size(inpt: torch.Tensor) -> list[int]:

From ac82cea8f5bb038578b0f46d4a40c0b6bd772ded Mon Sep 17 00:00:00 2001
From: Justin Davis <judavis@nvidia.com>
Date: Mon, 17 Nov 2025 16:14:16 -0800
Subject: [PATCH 08/18] draft initial gaussian_blur cvcuda kernel
 implementation

---
 test/test_transforms_v2.py                    | 64 +++++++++++++++++++
 .../transforms/v2/functional/__init__.py      |  1 +
 torchvision/transforms/v2/functional/_misc.py | 40 ++++++++++--
 3 files changed, 99 insertions(+), 6 deletions(-)

diff --git a/test/test_transforms_v2.py b/test/test_transforms_v2.py
index f767e211125..3e9cda02226 100644
--- a/test/test_transforms_v2.py
+++ b/test/test_transforms_v2.py
@@ -4048,6 +4048,70 @@ def test_functional_image_correctness(self, dimensions, kernel_size, sigma, dtyp
         torch.testing.assert_close(actual, expected, rtol=0, atol=1)
 
 
+@pytest.mark.skipif(not CVCUDA_AVAILABLE, reason="test requires CVCUDA")
+@needs_cuda
+class TestGaussianBlurCVCUDA:
+    def test_kernel_image_errors(self):
+        image = make_image_cvcuda(batch_dims=(1,))
+
+        with pytest.raises(ValueError, match="kernel_size is a sequence its length should be 2"):
+            F.gaussian_blur_cvcuda(image, kernel_size=[1, 2, 3])
+
+        for kernel_size in [2, -1]:
+            with pytest.raises(ValueError, match="kernel_size should have odd and positive integers"):
+                F.gaussian_blur_cvcuda(image, kernel_size=kernel_size)
+
+        with pytest.raises(ValueError, match="sigma is a sequence, its length should be 2"):
+            F.gaussian_blur_cvcuda(image, kernel_size=1, sigma=[1, 2, 3])
+
+        with pytest.raises(TypeError, match="sigma should be either float or sequence of floats"):
+            F.gaussian_blur_cvcuda(image, kernel_size=1, sigma=object())
+
+        with pytest.raises(ValueError, match="sigma should have positive values"):
+            F.gaussian_blur_cvcuda(image, kernel_size=1, sigma=-1)
+
+    def test_functional(self):
+        check_functional(F.gaussian_blur, make_image_cvcuda(batch_dims=(1,)), kernel_size=(3, 3))
+
+    @pytest.mark.parametrize("device", cpu_and_cuda())
+    @pytest.mark.parametrize("sigma", [5, 2.0, (0.5, 2), [1.3, 2.7]])
+    def test_transform(self, device, sigma):
+        check_transform(
+            transforms.GaussianBlur(kernel_size=3, sigma=sigma), make_image_cvcuda(batch_dims=(1,), device=device)
+        )
+
+    @pytest.mark.parametrize(
+        ("dimensions", "kernel_size", "sigma"),
+        [
+            ((10, 12), (3, 3), 0.8),
+            ((10, 12), (3, 3), 0.5),
+            ((10, 12), (3, 5), 0.8),
+            ((10, 12), (3, 5), 0.5),
+            ((26, 28), (23, 23), 1.7),
+        ],
+    )
+    @pytest.mark.parametrize("color_space", ["RGB", "GRAY"])
+    @pytest.mark.parametrize("batch_dims", [(1,), (2,), (4,)])
+    @pytest.mark.parametrize("dtype", [torch.uint8, torch.float32])
+    def test_functional_image_correctness(self, dimensions, kernel_size, sigma, color_space, batch_dims, dtype):
+        height, width = dimensions
+
+        image_tensor = make_image(
+            size=(height, width), color_space=color_space, batch_dims=batch_dims, dtype=dtype, device="cuda"
+        )
+        image_cvcuda = F.to_cvcuda_tensor(image_tensor)
+
+        expected = F.gaussian_blur_image(image_tensor, kernel_size=kernel_size, sigma=sigma)
+        actual = F.gaussian_blur_cvcuda(image_cvcuda, kernel_size=kernel_size, sigma=sigma)
+        actual_torch = F.cvcuda_to_tensor(actual)
+
+        if dtype.is_floating_point:
+            torch.testing.assert_close(actual_torch, expected, rtol=0, atol=0.3)
+        else:
+            # uint8/16 gaussians can differ by up to max-value, most likely an overflow issue
+            torch.testing.assert_close(actual_torch, expected, rtol=0, atol=get_max_value(dtype))
+
+
 class TestGaussianNoise:
     @pytest.mark.parametrize(
         "make_input",
diff --git a/torchvision/transforms/v2/functional/__init__.py b/torchvision/transforms/v2/functional/__init__.py
index 032a993b1f0..e948118a8bc 100644
--- a/torchvision/transforms/v2/functional/__init__.py
+++ b/torchvision/transforms/v2/functional/__init__.py
@@ -147,6 +147,7 @@
 from ._misc import (
     convert_image_dtype,
     gaussian_blur,
+    gaussian_blur_cvcuda,
     gaussian_blur_image,
     gaussian_blur_video,
     gaussian_noise,
diff --git a/torchvision/transforms/v2/functional/_misc.py b/torchvision/transforms/v2/functional/_misc.py
index 0fa05a2113c..eb3773e47d6 100644
--- a/torchvision/transforms/v2/functional/_misc.py
+++ b/torchvision/transforms/v2/functional/_misc.py
@@ -1,5 +1,5 @@
 import math
-from typing import Optional, TYPE_CHECKING
+from typing import Optional, Sequence, TYPE_CHECKING
 
 import PIL.Image
 import torch
@@ -106,11 +106,10 @@ def _get_gaussian_kernel2d(
     return kernel2d
 
 
-@_register_kernel_internal(gaussian_blur, torch.Tensor)
-@_register_kernel_internal(gaussian_blur, tv_tensors.Image)
-def gaussian_blur_image(
-    image: torch.Tensor, kernel_size: list[int], sigma: Optional[list[float]] = None
-) -> torch.Tensor:
+def _validate_kernel_size_and_sigma(
+    kernel_size: Sequence[int] | int,
+    sigma: Sequence[float | int] | float | int | None = None,
+) -> tuple[list[int], list[float]]:
     # TODO: consider deprecating integers from sigma on the future
     if isinstance(kernel_size, int):
         kernel_size = [kernel_size, kernel_size]
@@ -139,6 +138,16 @@ def gaussian_blur_image(
         if s <= 0.0:
             raise ValueError(f"sigma should have positive values. Got {sigma}")
 
+    return kernel_size, sigma
+
+
+@_register_kernel_internal(gaussian_blur, torch.Tensor)
+@_register_kernel_internal(gaussian_blur, tv_tensors.Image)
+def gaussian_blur_image(
+    image: torch.Tensor, kernel_size: list[int], sigma: Optional[list[float]] = None
+) -> torch.Tensor:
+    kernel_size, sigma = _validate_kernel_size_and_sigma(kernel_size, sigma)
+
     if image.numel() == 0:
         return image
 
@@ -188,6 +197,25 @@ def gaussian_blur_video(
     return gaussian_blur_image(video, kernel_size, sigma)
 
 
+def gaussian_blur_cvcuda(
+    image: "cvcuda.Tensor",
+    kernel_size: Sequence[int] | int,
+    sigma: Sequence[float | int] | float | int | None = None,
+) -> "cvcuda.Tensor":
+    kernel_size, sigma = _validate_kernel_size_and_sigma(kernel_size, sigma)
+
+    return cvcuda.gaussian(
+        image,
+        tuple(kernel_size),
+        tuple(sigma),
+        border=cvcuda.Border.REFLECT,
+    )
+
+
+if CVCUDA_AVAILABLE:
+    _gaussian_blur_cvcuda = _register_kernel_internal(gaussian_blur, cvcuda.Tensor)(gaussian_blur_cvcuda)
+
+
 def gaussian_noise(inpt: torch.Tensor, mean: float = 0.0, sigma: float = 0.1, clip: bool = True) -> torch.Tensor:
     """See :class:`~torchvision.transforms.v2.GaussianNoise`"""
     if torch.jit.is_scripting():

From b18fedf678c1975ecb39935b75297603514a1ed9 Mon Sep 17 00:00:00 2001
From: Justin Davis <judavis@nvidia.com>
Date: Thu, 20 Nov 2025 09:00:21 -0800
Subject: [PATCH 09/18] fix: move cvcuda tests to centralized class, more
 guards againist no import, explicit imports in func

---
 test/test_transforms_v2.py                    | 66 ++++++++-----------
 .../transforms/v2/functional/__init__.py      |  1 -
 torchvision/transforms/v2/functional/_misc.py | 42 ++++++++++--
 3 files changed, 64 insertions(+), 45 deletions(-)

diff --git a/test/test_transforms_v2.py b/test/test_transforms_v2.py
index 3e9cda02226..b08c2d3fcd5 100644
--- a/test/test_transforms_v2.py
+++ b/test/test_transforms_v2.py
@@ -3937,7 +3937,15 @@ def test_kernel_video(self):
 
     @pytest.mark.parametrize(
         "make_input",
-        [make_image_tensor, make_image_pil, make_image, make_video],
+        [
+            make_image_tensor,
+            make_image_pil,
+            make_image,
+            make_video,
+            pytest.param(
+                make_image_cvcuda, marks=pytest.mark.skipif(not CVCUDA_AVAILABLE, reason="CVCUDA is not available")
+            ),
+        ],
     )
     def test_functional(self, make_input):
         check_functional(F.gaussian_blur, make_input(), kernel_size=(3, 3))
@@ -3949,6 +3957,11 @@ def test_functional(self, make_input):
             (F._misc._gaussian_blur_image_pil, PIL.Image.Image),
             (F.gaussian_blur_image, tv_tensors.Image),
             (F.gaussian_blur_video, tv_tensors.Video),
+            pytest.param(
+                F._misc._gaussian_blur_cvcuda,
+                cvcuda.Tensor,
+                marks=pytest.mark.skipif(not CVCUDA_AVAILABLE, reason="CVCUDA is not available"),
+            ),
         ],
     )
     def test_functional_signature(self, kernel, input_type):
@@ -3956,7 +3969,17 @@ def test_functional_signature(self, kernel, input_type):
 
     @pytest.mark.parametrize(
         "make_input",
-        [make_image_tensor, make_image_pil, make_image, make_bounding_boxes, make_segmentation_mask, make_video],
+        [
+            make_image_tensor,
+            make_image_pil,
+            make_image,
+            make_bounding_boxes,
+            make_segmentation_mask,
+            make_video,
+            pytest.param(
+                make_image_cvcuda, marks=pytest.mark.skipif(not CVCUDA_AVAILABLE, reason="CVCUDA is not available")
+            ),
+        ],
     )
     @pytest.mark.parametrize("device", cpu_and_cuda())
     @pytest.mark.parametrize("sigma", [5, 2.0, (0.5, 2), [1.3, 2.7]])
@@ -4047,39 +4070,8 @@ def test_functional_image_correctness(self, dimensions, kernel_size, sigma, dtyp
 
         torch.testing.assert_close(actual, expected, rtol=0, atol=1)
 
-
-@pytest.mark.skipif(not CVCUDA_AVAILABLE, reason="test requires CVCUDA")
-@needs_cuda
-class TestGaussianBlurCVCUDA:
-    def test_kernel_image_errors(self):
-        image = make_image_cvcuda(batch_dims=(1,))
-
-        with pytest.raises(ValueError, match="kernel_size is a sequence its length should be 2"):
-            F.gaussian_blur_cvcuda(image, kernel_size=[1, 2, 3])
-
-        for kernel_size in [2, -1]:
-            with pytest.raises(ValueError, match="kernel_size should have odd and positive integers"):
-                F.gaussian_blur_cvcuda(image, kernel_size=kernel_size)
-
-        with pytest.raises(ValueError, match="sigma is a sequence, its length should be 2"):
-            F.gaussian_blur_cvcuda(image, kernel_size=1, sigma=[1, 2, 3])
-
-        with pytest.raises(TypeError, match="sigma should be either float or sequence of floats"):
-            F.gaussian_blur_cvcuda(image, kernel_size=1, sigma=object())
-
-        with pytest.raises(ValueError, match="sigma should have positive values"):
-            F.gaussian_blur_cvcuda(image, kernel_size=1, sigma=-1)
-
-    def test_functional(self):
-        check_functional(F.gaussian_blur, make_image_cvcuda(batch_dims=(1,)), kernel_size=(3, 3))
-
-    @pytest.mark.parametrize("device", cpu_and_cuda())
-    @pytest.mark.parametrize("sigma", [5, 2.0, (0.5, 2), [1.3, 2.7]])
-    def test_transform(self, device, sigma):
-        check_transform(
-            transforms.GaussianBlur(kernel_size=3, sigma=sigma), make_image_cvcuda(batch_dims=(1,), device=device)
-        )
-
+    @pytest.mark.skipif(not CVCUDA_AVAILABLE, reason="test requires CVCUDA")
+    @needs_cuda
     @pytest.mark.parametrize(
         ("dimensions", "kernel_size", "sigma"),
         [
@@ -4093,7 +4085,7 @@ def test_transform(self, device, sigma):
     @pytest.mark.parametrize("color_space", ["RGB", "GRAY"])
     @pytest.mark.parametrize("batch_dims", [(1,), (2,), (4,)])
     @pytest.mark.parametrize("dtype", [torch.uint8, torch.float32])
-    def test_functional_image_correctness(self, dimensions, kernel_size, sigma, color_space, batch_dims, dtype):
+    def test_functional_cvcuda_parity(self, dimensions, kernel_size, sigma, color_space, batch_dims, dtype):
         height, width = dimensions
 
         image_tensor = make_image(
@@ -4102,7 +4094,7 @@ def test_functional_image_correctness(self, dimensions, kernel_size, sigma, colo
         image_cvcuda = F.to_cvcuda_tensor(image_tensor)
 
         expected = F.gaussian_blur_image(image_tensor, kernel_size=kernel_size, sigma=sigma)
-        actual = F.gaussian_blur_cvcuda(image_cvcuda, kernel_size=kernel_size, sigma=sigma)
+        actual = F._misc._gaussian_blur_cvcuda(image_cvcuda, kernel_size=kernel_size, sigma=sigma)
         actual_torch = F.cvcuda_to_tensor(actual)
 
         if dtype.is_floating_point:
diff --git a/torchvision/transforms/v2/functional/__init__.py b/torchvision/transforms/v2/functional/__init__.py
index e948118a8bc..032a993b1f0 100644
--- a/torchvision/transforms/v2/functional/__init__.py
+++ b/torchvision/transforms/v2/functional/__init__.py
@@ -147,7 +147,6 @@
 from ._misc import (
     convert_image_dtype,
     gaussian_blur,
-    gaussian_blur_cvcuda,
     gaussian_blur_image,
     gaussian_blur_video,
     gaussian_noise,
diff --git a/torchvision/transforms/v2/functional/_misc.py b/torchvision/transforms/v2/functional/_misc.py
index eb3773e47d6..59b11da43f5 100644
--- a/torchvision/transforms/v2/functional/_misc.py
+++ b/torchvision/transforms/v2/functional/_misc.py
@@ -110,7 +110,7 @@ def _validate_kernel_size_and_sigma(
     kernel_size: Sequence[int] | int,
     sigma: Sequence[float | int] | float | int | None = None,
 ) -> tuple[list[int], list[float]]:
-    # TODO: consider deprecating integers from sigma on the future
+    # duplicated logic from gaussian_blur_image for use in gaussian_blur_cvcuda
     if isinstance(kernel_size, int):
         kernel_size = [kernel_size, kernel_size]
     elif len(kernel_size) != 2:
@@ -146,7 +146,33 @@ def _validate_kernel_size_and_sigma(
 def gaussian_blur_image(
     image: torch.Tensor, kernel_size: list[int], sigma: Optional[list[float]] = None
 ) -> torch.Tensor:
-    kernel_size, sigma = _validate_kernel_size_and_sigma(kernel_size, sigma)
+    # TODO: consider deprecating integers from sigma on the future
+    if isinstance(kernel_size, int):
+        kernel_size = [kernel_size, kernel_size]
+    elif len(kernel_size) != 2:
+        raise ValueError(f"If kernel_size is a sequence its length should be 2. Got {len(kernel_size)}")
+    for ksize in kernel_size:
+        if ksize % 2 == 0 or ksize < 0:
+            raise ValueError(f"kernel_size should have odd and positive integers. Got {kernel_size}")
+
+    if sigma is None:
+        sigma = [ksize * 0.15 + 0.35 for ksize in kernel_size]
+    else:
+        if isinstance(sigma, (list, tuple)):
+            length = len(sigma)
+            if length == 1:
+                s = sigma[0]
+                sigma = [s, s]
+            elif length != 2:
+                raise ValueError(f"If sigma is a sequence, its length should be 2. Got {length}")
+        elif isinstance(sigma, (int, float)):
+            s = float(sigma)
+            sigma = [s, s]
+        else:
+            raise TypeError(f"sigma should be either float or sequence of floats. Got {type(sigma)}")
+    for s in sigma:
+        if s <= 0.0:
+            raise ValueError(f"sigma should have positive values. Got {sigma}")
 
     if image.numel() == 0:
         return image
@@ -197,11 +223,11 @@ def gaussian_blur_video(
     return gaussian_blur_image(video, kernel_size, sigma)
 
 
-def gaussian_blur_cvcuda(
-    image: "cvcuda.Tensor",
-    kernel_size: Sequence[int] | int,
-    sigma: Sequence[float | int] | float | int | None = None,
+def _gaussian_blur_cvcuda(
+    image: "cvcuda.Tensor", kernel_size: list[int], sigma: Optional[list[float]] = None
 ) -> "cvcuda.Tensor":
+    cvcuda = _import_cvcuda()
+
     kernel_size, sigma = _validate_kernel_size_and_sigma(kernel_size, sigma)
 
     return cvcuda.gaussian(
@@ -213,7 +239,9 @@ def gaussian_blur_cvcuda(
 
 
 if CVCUDA_AVAILABLE:
-    _gaussian_blur_cvcuda = _register_kernel_internal(gaussian_blur, cvcuda.Tensor)(gaussian_blur_cvcuda)
+    _gaussian_blur_cvcuda_registered = _register_kernel_internal(gaussian_blur, _import_cvcuda().Tensor)(
+        _gaussian_blur_cvcuda
+    )
 
 
 def gaussian_noise(inpt: torch.Tensor, mean: float = 0.0, sigma: float = 0.1, clip: bool = True) -> torch.Tensor:

From 5df3a7d7c6be7b7f2712bd51741aa87afa935a28 Mon Sep 17 00:00:00 2001
From: Justin Davis <judavis@nvidia.com>
Date: Mon, 24 Nov 2025 12:37:53 -0800
Subject: [PATCH 10/18] consolidate gaussian_blur_image to use new
 validate_kernel_size_and_sigma

---
 torchvision/transforms/v2/functional/_misc.py | 36 +++----------------
 1 file changed, 5 insertions(+), 31 deletions(-)

diff --git a/torchvision/transforms/v2/functional/_misc.py b/torchvision/transforms/v2/functional/_misc.py
index 59b11da43f5..db5c86fa424 100644
--- a/torchvision/transforms/v2/functional/_misc.py
+++ b/torchvision/transforms/v2/functional/_misc.py
@@ -1,5 +1,5 @@
 import math
-from typing import Optional, Sequence, TYPE_CHECKING
+from typing import List, Optional, TYPE_CHECKING, Union
 
 import PIL.Image
 import torch
@@ -107,10 +107,10 @@ def _get_gaussian_kernel2d(
 
 
 def _validate_kernel_size_and_sigma(
-    kernel_size: Sequence[int] | int,
-    sigma: Sequence[float | int] | float | int | None = None,
+    kernel_size: List[int] | int,
+    sigma: Optional[Union[List[float], float, int]] = None,
 ) -> tuple[list[int], list[float]]:
-    # duplicated logic from gaussian_blur_image for use in gaussian_blur_cvcuda
+    # TODO: consider deprecating integers from sigma on the future
     if isinstance(kernel_size, int):
         kernel_size = [kernel_size, kernel_size]
     elif len(kernel_size) != 2:
@@ -146,33 +146,7 @@ def _validate_kernel_size_and_sigma(
 def gaussian_blur_image(
     image: torch.Tensor, kernel_size: list[int], sigma: Optional[list[float]] = None
 ) -> torch.Tensor:
-    # TODO: consider deprecating integers from sigma on the future
-    if isinstance(kernel_size, int):
-        kernel_size = [kernel_size, kernel_size]
-    elif len(kernel_size) != 2:
-        raise ValueError(f"If kernel_size is a sequence its length should be 2. Got {len(kernel_size)}")
-    for ksize in kernel_size:
-        if ksize % 2 == 0 or ksize < 0:
-            raise ValueError(f"kernel_size should have odd and positive integers. Got {kernel_size}")
-
-    if sigma is None:
-        sigma = [ksize * 0.15 + 0.35 for ksize in kernel_size]
-    else:
-        if isinstance(sigma, (list, tuple)):
-            length = len(sigma)
-            if length == 1:
-                s = sigma[0]
-                sigma = [s, s]
-            elif length != 2:
-                raise ValueError(f"If sigma is a sequence, its length should be 2. Got {length}")
-        elif isinstance(sigma, (int, float)):
-            s = float(sigma)
-            sigma = [s, s]
-        else:
-            raise TypeError(f"sigma should be either float or sequence of floats. Got {type(sigma)}")
-    for s in sigma:
-        if s <= 0.0:
-            raise ValueError(f"sigma should have positive values. Got {sigma}")
+    kernel_size, sigma = _validate_kernel_size_and_sigma(kernel_size, sigma)
 
     if image.numel() == 0:
         return image

From 9cd75823f7288a060bf8f9fc774c0336379feff6 Mon Sep 17 00:00:00 2001
From: Justin Davis <judavis@nvidia.com>
Date: Mon, 24 Nov 2025 12:49:47 -0800
Subject: [PATCH 11/18] resolve more review comments

---
 test/test_transforms_v2.py                    | 5 ++++-
 torchvision/transforms/v2/functional/_misc.py | 8 +++-----
 2 files changed, 7 insertions(+), 6 deletions(-)

diff --git a/test/test_transforms_v2.py b/test/test_transforms_v2.py
index b08c2d3fcd5..449708fd25f 100644
--- a/test/test_transforms_v2.py
+++ b/test/test_transforms_v2.py
@@ -4085,7 +4085,7 @@ def test_functional_image_correctness(self, dimensions, kernel_size, sigma, dtyp
     @pytest.mark.parametrize("color_space", ["RGB", "GRAY"])
     @pytest.mark.parametrize("batch_dims", [(1,), (2,), (4,)])
     @pytest.mark.parametrize("dtype", [torch.uint8, torch.float32])
-    def test_functional_cvcuda_parity(self, dimensions, kernel_size, sigma, color_space, batch_dims, dtype):
+    def test_functional_cvcuda_correctness(self, dimensions, kernel_size, sigma, color_space, batch_dims, dtype):
         height, width = dimensions
 
         image_tensor = make_image(
@@ -4098,6 +4098,9 @@ def test_functional_cvcuda_parity(self, dimensions, kernel_size, sigma, color_sp
         actual_torch = F.cvcuda_to_tensor(actual)
 
         if dtype.is_floating_point:
+            # floating point precision differences between torch and cvcuda are present
+            # observed greatest absolute error is 0.3
+            # most likely stemming from different implementation
             torch.testing.assert_close(actual_torch, expected, rtol=0, atol=0.3)
         else:
             # uint8/16 gaussians can differ by up to max-value, most likely an overflow issue
diff --git a/torchvision/transforms/v2/functional/_misc.py b/torchvision/transforms/v2/functional/_misc.py
index db5c86fa424..2888e242e23 100644
--- a/torchvision/transforms/v2/functional/_misc.py
+++ b/torchvision/transforms/v2/functional/_misc.py
@@ -1,5 +1,5 @@
 import math
-from typing import List, Optional, TYPE_CHECKING, Union
+from typing import List, Optional, Tuple, TYPE_CHECKING, Union
 
 import PIL.Image
 import torch
@@ -109,7 +109,7 @@ def _get_gaussian_kernel2d(
 def _validate_kernel_size_and_sigma(
     kernel_size: List[int] | int,
     sigma: Optional[Union[List[float], float, int]] = None,
-) -> tuple[list[int], list[float]]:
+) -> Tuple[List[int], List[float]]:
     # TODO: consider deprecating integers from sigma on the future
     if isinstance(kernel_size, int):
         kernel_size = [kernel_size, kernel_size]
@@ -213,9 +213,7 @@ def _gaussian_blur_cvcuda(
 
 
 if CVCUDA_AVAILABLE:
-    _gaussian_blur_cvcuda_registered = _register_kernel_internal(gaussian_blur, _import_cvcuda().Tensor)(
-        _gaussian_blur_cvcuda
-    )
+    _register_kernel_internal(gaussian_blur, _import_cvcuda().Tensor)(_gaussian_blur_cvcuda)
 
 
 def gaussian_noise(inpt: torch.Tensor, mean: float = 0.0, sigma: float = 0.1, clip: bool = True) -> torch.Tensor:

From d9e3f839c2f1ccadfbf3b4d3e8db28514279d069 Mon Sep 17 00:00:00 2001
From: Justin Davis <judavis@nvidia.com>
Date: Mon, 24 Nov 2025 12:52:24 -0800
Subject: [PATCH 12/18] match type hint in validate to gaussian_blur_image

---
 torchvision/transforms/v2/functional/_misc.py | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/torchvision/transforms/v2/functional/_misc.py b/torchvision/transforms/v2/functional/_misc.py
index 2888e242e23..877898605f1 100644
--- a/torchvision/transforms/v2/functional/_misc.py
+++ b/torchvision/transforms/v2/functional/_misc.py
@@ -1,5 +1,5 @@
 import math
-from typing import List, Optional, Tuple, TYPE_CHECKING, Union
+from typing import List, Optional, Tuple, TYPE_CHECKING
 
 import PIL.Image
 import torch
@@ -107,8 +107,8 @@ def _get_gaussian_kernel2d(
 
 
 def _validate_kernel_size_and_sigma(
-    kernel_size: List[int] | int,
-    sigma: Optional[Union[List[float], float, int]] = None,
+    kernel_size: List[int],
+    sigma: Optional[List[float]] = None,
 ) -> Tuple[List[int], List[float]]:
     # TODO: consider deprecating integers from sigma on the future
     if isinstance(kernel_size, int):

From ccd4c1d04d19c057f2bec4f33624d732dd85c180 Mon Sep 17 00:00:00 2001
From: Justin Davis <judavis@nvidia.com>
Date: Tue, 25 Nov 2025 16:00:40 -0800
Subject: [PATCH 13/18] resolve tests failing due to cvcuda.Tensor

---
 test/test_transforms_v2.py | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/test/test_transforms_v2.py b/test/test_transforms_v2.py
index 449708fd25f..5ccb2fcee4e 100644
--- a/test/test_transforms_v2.py
+++ b/test/test_transforms_v2.py
@@ -3959,12 +3959,14 @@ def test_functional(self, make_input):
             (F.gaussian_blur_video, tv_tensors.Video),
             pytest.param(
                 F._misc._gaussian_blur_cvcuda,
-                cvcuda.Tensor,
+                "cvcuda.Tensor",
                 marks=pytest.mark.skipif(not CVCUDA_AVAILABLE, reason="CVCUDA is not available"),
             ),
         ],
     )
     def test_functional_signature(self, kernel, input_type):
+        if input_type == "cvcuda.Tensor":
+            input_type = _import_cvcuda().Tensor
         check_functional_kernel_signature_match(F.gaussian_blur, kernel=kernel, input_type=input_type)
 
     @pytest.mark.parametrize(

From b3d7814c869496a35a117566963ee49389cb1a4c Mon Sep 17 00:00:00 2001
From: Justin Davis <judavis@nvidia.com>
Date: Wed, 26 Nov 2025 09:39:20 -0800
Subject: [PATCH 14/18] fix guassian border mode to adhere to torch/opencv

---
 test/test_transforms_v2.py                    | 59 +++++++------------
 torchvision/transforms/v2/functional/_misc.py |  2 +-
 2 files changed, 22 insertions(+), 39 deletions(-)

diff --git a/test/test_transforms_v2.py b/test/test_transforms_v2.py
index 5ccb2fcee4e..1ec7c0fb99b 100644
--- a/test/test_transforms_v2.py
+++ b/test/test_transforms_v2.py
@@ -4044,11 +4044,22 @@ def test_make_params(self, sigma):
             ((1, 26, 28), (23, 23), 1.7),
         ],
     )
-    @pytest.mark.parametrize("dtype", [torch.float32, torch.float64, torch.float16])
+    @pytest.mark.parametrize("dtype", [torch.uint8, torch.float32, torch.float64, torch.float16])
     @pytest.mark.parametrize("device", cpu_and_cuda())
-    def test_functional_image_correctness(self, dimensions, kernel_size, sigma, dtype, device):
+    @pytest.mark.parametrize(
+        "input_type",
+        [
+            tv_tensors.Image,
+            pytest.param(
+                "cvcuda.Tensor", marks=pytest.mark.skipif(not CVCUDA_AVAILABLE, reason="CVCUDA not available")
+            ),
+        ],
+    )
+    def test_functional_image_correctness(self, dimensions, kernel_size, sigma, dtype, device, input_type):
         if dtype is torch.float16 and device == "cpu":
             pytest.skip("The CPU implementation of float16 on CPU differs from opencv")
+        if (dtype != torch.float32 and dtype != torch.uint8) and input_type == "cvcuda.Tensor":
+            pytest.skip("CVCUDA does not support non-float32 or uint8 dtypes for gaussian blur")
 
         num_channels, height, width = dimensions
 
@@ -4068,45 +4079,17 @@ def test_functional_image_correctness(self, dimensions, kernel_size, sigma, dtyp
             device=device,
         )
 
-        actual = F.gaussian_blur_image(image, kernel_size=kernel_size, sigma=sigma)
-
-        torch.testing.assert_close(actual, expected, rtol=0, atol=1)
-
-    @pytest.mark.skipif(not CVCUDA_AVAILABLE, reason="test requires CVCUDA")
-    @needs_cuda
-    @pytest.mark.parametrize(
-        ("dimensions", "kernel_size", "sigma"),
-        [
-            ((10, 12), (3, 3), 0.8),
-            ((10, 12), (3, 3), 0.5),
-            ((10, 12), (3, 5), 0.8),
-            ((10, 12), (3, 5), 0.5),
-            ((26, 28), (23, 23), 1.7),
-        ],
-    )
-    @pytest.mark.parametrize("color_space", ["RGB", "GRAY"])
-    @pytest.mark.parametrize("batch_dims", [(1,), (2,), (4,)])
-    @pytest.mark.parametrize("dtype", [torch.uint8, torch.float32])
-    def test_functional_cvcuda_correctness(self, dimensions, kernel_size, sigma, color_space, batch_dims, dtype):
-        height, width = dimensions
+        if input_type == "cvcuda.Tensor":
+            image = image.unsqueeze(0)
+            image = F.to_cvcuda_tensor(image)
 
-        image_tensor = make_image(
-            size=(height, width), color_space=color_space, batch_dims=batch_dims, dtype=dtype, device="cuda"
-        )
-        image_cvcuda = F.to_cvcuda_tensor(image_tensor)
+        actual = F.gaussian_blur(image, kernel_size=kernel_size, sigma=sigma)
 
-        expected = F.gaussian_blur_image(image_tensor, kernel_size=kernel_size, sigma=sigma)
-        actual = F._misc._gaussian_blur_cvcuda(image_cvcuda, kernel_size=kernel_size, sigma=sigma)
-        actual_torch = F.cvcuda_to_tensor(actual)
+        if input_type == "cvcuda.Tensor":
+            actual = F.cvcuda_to_tensor(actual)
+            actual = actual.squeeze(0).to(device=device)
 
-        if dtype.is_floating_point:
-            # floating point precision differences between torch and cvcuda are present
-            # observed greatest absolute error is 0.3
-            # most likely stemming from different implementation
-            torch.testing.assert_close(actual_torch, expected, rtol=0, atol=0.3)
-        else:
-            # uint8/16 gaussians can differ by up to max-value, most likely an overflow issue
-            torch.testing.assert_close(actual_torch, expected, rtol=0, atol=get_max_value(dtype))
+        torch.testing.assert_close(actual, expected, rtol=0, atol=1)
 
 
 class TestGaussianNoise:
diff --git a/torchvision/transforms/v2/functional/_misc.py b/torchvision/transforms/v2/functional/_misc.py
index 877898605f1..dbb10b02e8d 100644
--- a/torchvision/transforms/v2/functional/_misc.py
+++ b/torchvision/transforms/v2/functional/_misc.py
@@ -208,7 +208,7 @@ def _gaussian_blur_cvcuda(
         image,
         tuple(kernel_size),
         tuple(sigma),
-        border=cvcuda.Border.REFLECT,
+        border=cvcuda.Border.REFLECT101,
     )
 
 

From 160e047172c1f0de025a050e38f0f4d9d855045f Mon Sep 17 00:00:00 2001
From: Justin Davis <judavis@nvidia.com>
Date: Mon, 1 Dec 2025 18:34:11 -0800
Subject: [PATCH 15/18] remove unneeded cvcuda setup

---
 torchvision/transforms/v2/functional/_augment.py  | 11 +----------
 torchvision/transforms/v2/functional/_color.py    | 12 +-----------
 torchvision/transforms/v2/functional/_geometry.py |  2 +-
 3 files changed, 3 insertions(+), 22 deletions(-)

diff --git a/torchvision/transforms/v2/functional/_augment.py b/torchvision/transforms/v2/functional/_augment.py
index 7ce5bdc7b7e..a904d8d7cbd 100644
--- a/torchvision/transforms/v2/functional/_augment.py
+++ b/torchvision/transforms/v2/functional/_augment.py
@@ -1,5 +1,4 @@
 import io
-from typing import TYPE_CHECKING
 
 import PIL.Image
 
@@ -9,15 +8,7 @@
 from torchvision.transforms.functional import pil_to_tensor, to_pil_image
 from torchvision.utils import _log_api_usage_once
 
-from ._utils import _get_kernel, _import_cvcuda, _is_cvcuda_available, _register_kernel_internal
-
-
-CVCUDA_AVAILABLE = _is_cvcuda_available()
-
-if TYPE_CHECKING:
-    import cvcuda  # type: ignore[import-not-found]
-if CVCUDA_AVAILABLE:
-    cvcuda = _import_cvcuda()  # noqa: F811
+from ._utils import _get_kernel, _register_kernel_internal
 
 
 def erase(
diff --git a/torchvision/transforms/v2/functional/_color.py b/torchvision/transforms/v2/functional/_color.py
index 5be9c62902a..be254c0d63a 100644
--- a/torchvision/transforms/v2/functional/_color.py
+++ b/torchvision/transforms/v2/functional/_color.py
@@ -1,5 +1,3 @@
-from typing import TYPE_CHECKING
-
 import PIL.Image
 import torch
 from torch.nn.functional import conv2d
@@ -11,15 +9,7 @@
 
 from ._misc import _num_value_bits, to_dtype_image
 from ._type_conversion import pil_to_tensor, to_pil_image
-from ._utils import _get_kernel, _import_cvcuda, _is_cvcuda_available, _register_kernel_internal
-
-
-CVCUDA_AVAILABLE = _is_cvcuda_available()
-
-if TYPE_CHECKING:
-    import cvcuda  # type: ignore[import-not-found]
-if CVCUDA_AVAILABLE:
-    cvcuda = _import_cvcuda()  # noqa: F811
+from ._utils import _get_kernel, _register_kernel_internal
 
 
 def rgb_to_grayscale(inpt: torch.Tensor, num_output_channels: int = 1) -> torch.Tensor:
diff --git a/torchvision/transforms/v2/functional/_geometry.py b/torchvision/transforms/v2/functional/_geometry.py
index 0e27218bc89..cb3ac70205a 100644
--- a/torchvision/transforms/v2/functional/_geometry.py
+++ b/torchvision/transforms/v2/functional/_geometry.py
@@ -2,7 +2,7 @@
 import numbers
 import warnings
 from collections.abc import Sequence
-from typing import Any, Optional, TYPE_CHECKING, Union
+from typing import Any, Optional, Union, TYPE_CHECKING
 
 import PIL.Image
 import torch

From 5ce83b11bfef26f04cfb1b2417b6877a6762197e Mon Sep 17 00:00:00 2001
From: Justin Davis <judavis@nvidia.com>
Date: Mon, 1 Dec 2025 18:35:16 -0800
Subject: [PATCH 16/18] use assert_close

---
 test/test_transforms_v2.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/test/test_transforms_v2.py b/test/test_transforms_v2.py
index 1ec7c0fb99b..0f89643f65c 100644
--- a/test/test_transforms_v2.py
+++ b/test/test_transforms_v2.py
@@ -4089,7 +4089,7 @@ def test_functional_image_correctness(self, dimensions, kernel_size, sigma, dtyp
             actual = F.cvcuda_to_tensor(actual)
             actual = actual.squeeze(0).to(device=device)
 
-        torch.testing.assert_close(actual, expected, rtol=0, atol=1)
+        assert_close(actual, expected, rtol=0, atol=1)
 
 
 class TestGaussianNoise:

From 3bcc517569949c918569fdc619228fcdfe491a13 Mon Sep 17 00:00:00 2001
From: Justin Davis <judavis@nvidia.com>
Date: Thu, 4 Dec 2025 11:14:51 -0800
Subject: [PATCH 17/18] update gaussian blur with main standards

---
 test/test_transforms_v2.py                    | 3 +--
 torchvision/transforms/v2/_misc.py            | 6 ++++++
 torchvision/transforms/v2/_utils.py           | 8 ++++----
 torchvision/transforms/v2/functional/_misc.py | 4 ++--
 4 files changed, 13 insertions(+), 8 deletions(-)

diff --git a/test/test_transforms_v2.py b/test/test_transforms_v2.py
index 0f89643f65c..8bf2bfa495d 100644
--- a/test/test_transforms_v2.py
+++ b/test/test_transforms_v2.py
@@ -25,7 +25,6 @@
     assert_equal,
     cache,
     cpu_and_cuda,
-    cvcuda_to_pil_compatible_tensor,
     freeze_rng_state,
     ignore_jit_no_profile_information_warning,
     make_bounding_boxes,
@@ -3958,7 +3957,7 @@ def test_functional(self, make_input):
             (F.gaussian_blur_image, tv_tensors.Image),
             (F.gaussian_blur_video, tv_tensors.Video),
             pytest.param(
-                F._misc._gaussian_blur_cvcuda,
+                F._misc._gaussian_blur_image_cvcuda,
                 "cvcuda.Tensor",
                 marks=pytest.mark.skipif(not CVCUDA_AVAILABLE, reason="CVCUDA is not available"),
             ),
diff --git a/torchvision/transforms/v2/_misc.py b/torchvision/transforms/v2/_misc.py
index 305149c87b1..c09e04b0834 100644
--- a/torchvision/transforms/v2/_misc.py
+++ b/torchvision/transforms/v2/_misc.py
@@ -9,6 +9,7 @@
 
 from torchvision import transforms as _transforms, tv_tensors
 from torchvision.transforms.v2 import functional as F, Transform
+from torchvision.transforms.v2.functional._utils import _is_cvcuda_available, _is_cvcuda_tensor
 
 from ._utils import (
     _parse_labels_getter,
@@ -20,6 +21,8 @@
     is_pure_tensor,
 )
 
+CVCUDA_AVAILABLE = _is_cvcuda_available()
+
 
 # TODO: do we want/need to expose this?
 class Identity(Transform):
@@ -192,6 +195,9 @@ class GaussianBlur(Transform):
 
     _v1_transform_cls = _transforms.GaussianBlur
 
+    if CVCUDA_AVAILABLE:
+        _transformed_types = Transform._transformed_types + (_is_cvcuda_tensor,)
+
     def __init__(
         self, kernel_size: Union[int, Sequence[int]], sigma: Union[int, float, Sequence[float]] = (0.1, 2.0)
     ) -> None:
diff --git a/torchvision/transforms/v2/_utils.py b/torchvision/transforms/v2/_utils.py
index 3fc33ce5964..e803aa49c60 100644
--- a/torchvision/transforms/v2/_utils.py
+++ b/torchvision/transforms/v2/_utils.py
@@ -15,8 +15,8 @@
 from torchvision._utils import sequence_to_str
 
 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_cvcuda_tensor, is_pure_tensor
-from torchvision.transforms.v2.functional._utils import _FillType, _FillTypeJIT
+from torchvision.transforms.v2.functional import get_dimensions, get_size, is_pure_tensor
+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, is_cvcuda_tensor))
+        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")
@@ -207,7 +207,7 @@ def query_size(flat_inputs: list[Any]) -> tuple[int, int]:
                 tv_tensors.Mask,
                 tv_tensors.BoundingBoxes,
                 tv_tensors.KeyPoints,
-                is_cvcuda_tensor,
+                _is_cvcuda_tensor,
             ),
         )
     }
diff --git a/torchvision/transforms/v2/functional/_misc.py b/torchvision/transforms/v2/functional/_misc.py
index dbb10b02e8d..1a7a3b53ac8 100644
--- a/torchvision/transforms/v2/functional/_misc.py
+++ b/torchvision/transforms/v2/functional/_misc.py
@@ -197,7 +197,7 @@ def gaussian_blur_video(
     return gaussian_blur_image(video, kernel_size, sigma)
 
 
-def _gaussian_blur_cvcuda(
+def _gaussian_blur_image_cvcuda(
     image: "cvcuda.Tensor", kernel_size: list[int], sigma: Optional[list[float]] = None
 ) -> "cvcuda.Tensor":
     cvcuda = _import_cvcuda()
@@ -213,7 +213,7 @@ def _gaussian_blur_cvcuda(
 
 
 if CVCUDA_AVAILABLE:
-    _register_kernel_internal(gaussian_blur, _import_cvcuda().Tensor)(_gaussian_blur_cvcuda)
+    _register_kernel_internal(gaussian_blur, _import_cvcuda().Tensor)(_gaussian_blur_image_cvcuda)
 
 
 def gaussian_noise(inpt: torch.Tensor, mean: float = 0.0, sigma: float = 0.1, clip: bool = True) -> torch.Tensor:

From 2edfdffb425e933d2e5e579d68b7068b643863dd Mon Sep 17 00:00:00 2001
From: Justin Davis <judavis@nvidia.com>
Date: Thu, 4 Dec 2025 13:44:54 -0800
Subject: [PATCH 18/18] check input type on kernel for signature test

---
 test/test_transforms_v2.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/test/test_transforms_v2.py b/test/test_transforms_v2.py
index 8bf2bfa495d..a5e394ec2ba 100644
--- a/test/test_transforms_v2.py
+++ b/test/test_transforms_v2.py
@@ -3958,13 +3958,13 @@ def test_functional(self, make_input):
             (F.gaussian_blur_video, tv_tensors.Video),
             pytest.param(
                 F._misc._gaussian_blur_image_cvcuda,
-                "cvcuda.Tensor",
+                None,
                 marks=pytest.mark.skipif(not CVCUDA_AVAILABLE, reason="CVCUDA is not available"),
             ),
         ],
     )
     def test_functional_signature(self, kernel, input_type):
-        if input_type == "cvcuda.Tensor":
+        if kernel is F._misc._gaussian_blur_image_cvcuda:
             input_type = _import_cvcuda().Tensor
         check_functional_kernel_signature_match(F.gaussian_blur, kernel=kernel, input_type=input_type)