From 44db71c0772e5ef5758c38d0e4e8ad9995946c80 Mon Sep 17 00:00:00 2001 From: Justin Davis Date: Tue, 25 Nov 2025 09:14:49 -0800 Subject: [PATCH 01/15] 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 Date: Tue, 25 Nov 2025 09:26:19 -0800 Subject: [PATCH 02/15] 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 Date: Mon, 1 Dec 2025 18:16:27 -0800 Subject: [PATCH 03/15] 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 Date: Mon, 1 Dec 2025 18:25:09 -0800 Subject: [PATCH 04/15] 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 Date: Tue, 2 Dec 2025 12:37:03 -0800 Subject: [PATCH 05/15] 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 Date: Tue, 2 Dec 2025 12:46:23 -0800 Subject: [PATCH 06/15] 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 Date: Thu, 4 Dec 2025 11:07:08 -0800 Subject: [PATCH 07/15] 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 fbea584365311ae6b56be7e4f6bbff1f834dd31a Mon Sep 17 00:00:00 2001 From: Justin Davis Date: Thu, 4 Dec 2025 11:15:49 -0800 Subject: [PATCH 08/15] update is cvcuda tensor impl --- torchvision/transforms/v2/_utils.py | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) 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, ), ) } From f59e9e9374dc91082d90d89027a1a84157d4b217 Mon Sep 17 00:00:00 2001 From: Justin Davis Date: Mon, 24 Nov 2025 17:01:24 -0800 Subject: [PATCH 09/15] begin work on pad --- test/test_transforms_v2.py | 9 ++++ .../transforms/v2/functional/_geometry.py | 43 +++++++++++++++++++ 2 files changed, 52 insertions(+) diff --git a/test/test_transforms_v2.py b/test/test_transforms_v2.py index f767e211125..864bd6772f0 100644 --- a/test/test_transforms_v2.py +++ b/test/test_transforms_v2.py @@ -4728,6 +4728,7 @@ def test_kernel_video(self): make_segmentation_mask, make_video, make_keypoints, + make_image_cvcuda, ], ) def test_functional(self, make_input): @@ -4746,9 +4747,16 @@ 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.pad_image_cvcuda, + "cvcuda.Tensor", + marks=pytest.mark.skipif(not CVCUDA_AVAILABLE, reason="test requires CVCUDA"), + ), ], ) def test_functional_signature(self, kernel, input_type): + if input_type == "cvcuda.Tensor": + input_type = _import_cvcuda().Tensor check_functional_kernel_signature_match(F.pad, kernel=kernel, input_type=input_type) @pytest.mark.parametrize( @@ -4761,6 +4769,7 @@ def test_functional_signature(self, kernel, input_type): make_segmentation_mask, make_video, make_keypoints, + make_image_cvcuda, ], ) def test_transform(self, make_input): diff --git a/torchvision/transforms/v2/functional/_geometry.py b/torchvision/transforms/v2/functional/_geometry.py index 0e27218bc89..c20e5796f7e 100644 --- a/torchvision/transforms/v2/functional/_geometry.py +++ b/torchvision/transforms/v2/functional/_geometry.py @@ -1682,6 +1682,49 @@ def _pad_with_vector_fill( _pad_image_pil = _register_kernel_internal(pad, PIL.Image.Image)(_FP.pad) +if _CVCUDA_AVAILABLE: + cvcuda = _import_cvcuda() + _pad_mode_to_cvcuda = { + "constant": cvcuda.BorderType.CONSTANT, + "reflect": cvcuda.BorderType.REFLECT, + "replicate": cvcuda.BorderType.REPLICATE, + "symmetric": cvcuda.BorderType.WRAP, + } + + +def _pad_cvcuda( + image: "cvcuda.Tensor", + padding: list[int], + fill: Optional[Union[int, float, list[float]]] = None, + padding_mode: str = "constant", +) -> "cvcuda.Tensor": + cvcuda = _import_cvcuda() + + if padding_mode not in _pad_mode_to_cvcuda: + raise ValueError(f"Padding mode '{padding_mode}' is not supported with CVCUDA") + + 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=_pad_mode_to_cvcuda[padding_mode], + border_value=fill, + top=top, + left=left, + bottom=bottom, + right=right, + ) + + +if _CVCUDA_AVAILABLE: + _register_kernel_internal(pad, _import_cvcuda().Tensor)(_pad_cvcuda) + + @_register_kernel_internal(pad, tv_tensors.Mask) def pad_mask( mask: torch.Tensor, From c799674b81f38eec71aafd8ee31e5f9adb4dc609 Mon Sep 17 00:00:00 2001 From: Justin Davis Date: Tue, 25 Nov 2025 09:00:36 -0800 Subject: [PATCH 10/15] pad passing all tests --- test/test_transforms_v2.py | 22 ++++++++++++++++++- .../transforms/v2/functional/_geometry.py | 12 +++++----- 2 files changed, 27 insertions(+), 7 deletions(-) diff --git a/test/test_transforms_v2.py b/test/test_transforms_v2.py index 864bd6772f0..f3895e16a00 100644 --- a/test/test_transforms_v2.py +++ b/test/test_transforms_v2.py @@ -4748,7 +4748,7 @@ def test_functional(self, make_input): (F.pad_mask, tv_tensors.Mask), (F.pad_video, tv_tensors.Video), pytest.param( - F.pad_image_cvcuda, + F._geometry._pad_cvcuda, "cvcuda.Tensor", marks=pytest.mark.skipif(not CVCUDA_AVAILABLE, reason="test requires CVCUDA"), ), @@ -4813,6 +4813,26 @@ def test_image_correctness(self, padding, padding_mode, fill, fn): assert_equal(actual, expected) + @pytest.mark.skipif(not CVCUDA_AVAILABLE, reason="test requires CVCUDA") + @pytest.mark.parametrize("padding", CORRECTNESS_PADDINGS) + @pytest.mark.parametrize( + ("padding_mode", "fill"), + [ + *[("constant", fill) for fill in CORRECTNESS_FILLS], + *[(padding_mode, None) for padding_mode in ["symmetric", "edge", "reflect"]], + ], + ) + @pytest.mark.parametrize("fn", [F.pad, transform_cls_to_functional(transforms.Pad)]) + def test_cvcuda_correctness(self, padding, padding_mode, fill, fn): + image = make_image_cvcuda(dtype=torch.uint8, device="cuda") + + fill = adapt_fill(fill, dtype=torch.uint8) + + actual = fn(image, padding=padding, padding_mode=padding_mode, fill=fill) + expected = F.pad(F.cvcuda_to_tensor(image), padding=padding, padding_mode=padding_mode, fill=fill) + + assert_equal(F.cvcuda_to_tensor(actual), expected) + def _reference_pad_bounding_boxes(self, bounding_boxes, *, padding): if isinstance(padding, int): padding = [padding] diff --git a/torchvision/transforms/v2/functional/_geometry.py b/torchvision/transforms/v2/functional/_geometry.py index c20e5796f7e..b50a93a7632 100644 --- a/torchvision/transforms/v2/functional/_geometry.py +++ b/torchvision/transforms/v2/functional/_geometry.py @@ -1683,12 +1683,12 @@ def _pad_with_vector_fill( if _CVCUDA_AVAILABLE: - cvcuda = _import_cvcuda() _pad_mode_to_cvcuda = { - "constant": cvcuda.BorderType.CONSTANT, - "reflect": cvcuda.BorderType.REFLECT, - "replicate": cvcuda.BorderType.REPLICATE, - "symmetric": cvcuda.BorderType.WRAP, + "constant": cvcuda.Border.CONSTANT, + "reflect": cvcuda.Border.REFLECT101, + "replicate": cvcuda.Border.REPLICATE, + "edge": cvcuda.Border.REPLICATE, + "symmetric": cvcuda.Border.REFLECT, } @@ -1700,7 +1700,7 @@ def _pad_cvcuda( ) -> "cvcuda.Tensor": cvcuda = _import_cvcuda() - if padding_mode not in _pad_mode_to_cvcuda: + if _pad_mode_to_cvcuda.get(padding_mode) is None: raise ValueError(f"Padding mode '{padding_mode}' is not supported with CVCUDA") if fill is None: From e2b0b3d911a34474ed446cb811e256c4b9f0788b Mon Sep 17 00:00:00 2001 From: Justin Davis Date: Mon, 1 Dec 2025 18:57:21 -0800 Subject: [PATCH 11/15] update to new setup from PR reviews --- test/test_transforms_v2.py | 41 +++++++++---------- torchvision/transforms/v2/_geometry.py | 3 ++ .../transforms/v2/functional/_geometry.py | 4 +- 3 files changed, 24 insertions(+), 24 deletions(-) diff --git a/test/test_transforms_v2.py b/test/test_transforms_v2.py index f3895e16a00..4867cf7809b 100644 --- a/test/test_transforms_v2.py +++ b/test/test_transforms_v2.py @@ -4728,7 +4728,9 @@ def test_kernel_video(self): make_segmentation_mask, make_video, make_keypoints, - make_image_cvcuda, + pytest.param( + make_image_cvcuda, marks=pytest.mark.skipif(not CVCUDA_AVAILABLE, reason="test requires CVCUDA") + ), ], ) def test_functional(self, make_input): @@ -4769,7 +4771,9 @@ def test_functional_signature(self, kernel, input_type): make_segmentation_mask, make_video, make_keypoints, - make_image_cvcuda, + pytest.param( + make_image_cvcuda, marks=pytest.mark.skipif(not CVCUDA_AVAILABLE, reason="test requires CVCUDA") + ), ], ) def test_transform(self, make_input): @@ -4794,26 +4798,15 @@ def test_transform_errors(self): with pytest.raises(ValueError, match="Padding mode should be either"): transforms.Pad(12, padding_mode="abc") - @pytest.mark.parametrize("padding", CORRECTNESS_PADDINGS) @pytest.mark.parametrize( - ("padding_mode", "fill"), + "make_input", [ - *[("constant", fill) for fill in CORRECTNESS_FILLS], - *[(padding_mode, None) for padding_mode in ["symmetric", "edge", "reflect"]], + make_image, + pytest.param( + make_image_cvcuda, marks=pytest.mark.skipif(not CVCUDA_AVAILABLE, reason="test requires CVCUDA") + ), ], ) - @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") - - fill = adapt_fill(fill, dtype=torch.uint8) - - actual = fn(image, padding=padding, padding_mode=padding_mode, fill=fill) - expected = F.to_image(F.pad(F.to_pil_image(image), padding=padding, padding_mode=padding_mode, fill=fill)) - - assert_equal(actual, expected) - - @pytest.mark.skipif(not CVCUDA_AVAILABLE, reason="test requires CVCUDA") @pytest.mark.parametrize("padding", CORRECTNESS_PADDINGS) @pytest.mark.parametrize( ("padding_mode", "fill"), @@ -4823,15 +4816,19 @@ def test_image_correctness(self, padding, padding_mode, fill, fn): ], ) @pytest.mark.parametrize("fn", [F.pad, transform_cls_to_functional(transforms.Pad)]) - def test_cvcuda_correctness(self, padding, padding_mode, fill, fn): - image = make_image_cvcuda(dtype=torch.uint8, device="cuda") + 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) - expected = F.pad(F.cvcuda_to_tensor(image), padding=padding, padding_mode=padding_mode, fill=fill) - assert_equal(F.cvcuda_to_tensor(actual), expected) + if make_input is make_image_cvcuda: + image = cvcuda_to_pil_compatible_tensor(image) + + expected = F.to_image(F.pad(F.to_pil_image(image), padding=padding, padding_mode=padding_mode, fill=fill)) + + assert_equal(actual, expected) def _reference_pad_bounding_boxes(self, bounding_boxes, *, padding): if isinstance(padding, int): diff --git a/torchvision/transforms/v2/_geometry.py b/torchvision/transforms/v2/_geometry.py index 96166e05e9a..02c1603a295 100644 --- a/torchvision/transforms/v2/_geometry.py +++ b/torchvision/transforms/v2/_geometry.py @@ -26,6 +26,7 @@ get_bounding_boxes, has_all, has_any, + is_cvcuda_tensor, is_pure_tensor, query_size, ) @@ -465,6 +466,8 @@ class Pad(Transform): _v1_transform_cls = _transforms.Pad + _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() diff --git a/torchvision/transforms/v2/functional/_geometry.py b/torchvision/transforms/v2/functional/_geometry.py index b50a93a7632..2463d130974 100644 --- a/torchvision/transforms/v2/functional/_geometry.py +++ b/torchvision/transforms/v2/functional/_geometry.py @@ -1682,7 +1682,7 @@ def _pad_with_vector_fill( _pad_image_pil = _register_kernel_internal(pad, PIL.Image.Image)(_FP.pad) -if _CVCUDA_AVAILABLE: +if CVCUDA_AVAILABLE: _pad_mode_to_cvcuda = { "constant": cvcuda.Border.CONSTANT, "reflect": cvcuda.Border.REFLECT101, @@ -1721,7 +1721,7 @@ def _pad_cvcuda( ) -if _CVCUDA_AVAILABLE: +if CVCUDA_AVAILABLE: _register_kernel_internal(pad, _import_cvcuda().Tensor)(_pad_cvcuda) From 95dcf5da0d67231ec4e42334aa3bc628c8aad67e Mon Sep 17 00:00:00 2001 From: Justin Davis Date: Thu, 4 Dec 2025 13:40:49 -0800 Subject: [PATCH 12/15] update pad to main standards --- test/test_transforms_v2.py | 9 +++--- torchvision/transforms/v2/_geometry.py | 4 +-- .../transforms/v2/functional/_geometry.py | 20 ++++--------- .../transforms/v2/functional/_utils.py | 28 ++++++++++++++++++- 4 files changed, 38 insertions(+), 23 deletions(-) diff --git a/test/test_transforms_v2.py b/test/test_transforms_v2.py index 4867cf7809b..abe5f3b6ab1 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, @@ -4750,14 +4749,14 @@ def test_functional(self, make_input): (F.pad_mask, tv_tensors.Mask), (F.pad_video, tv_tensors.Video), pytest.param( - F._geometry._pad_cvcuda, - "cvcuda.Tensor", + 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 input_type == "cvcuda.Tensor": + 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) @@ -4824,7 +4823,7 @@ def test_image_correctness(self, make_input, padding, padding_mode, fill, fn): actual = fn(image, padding=padding, padding_mode=padding_mode, fill=fill) if make_input is make_image_cvcuda: - image = cvcuda_to_pil_compatible_tensor(image) + 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)) diff --git a/torchvision/transforms/v2/_geometry.py b/torchvision/transforms/v2/_geometry.py index 02c1603a295..ee573a1374e 100644 --- a/torchvision/transforms/v2/_geometry.py +++ b/torchvision/transforms/v2/_geometry.py @@ -26,7 +26,6 @@ get_bounding_boxes, has_all, has_any, - is_cvcuda_tensor, is_pure_tensor, query_size, ) @@ -466,7 +465,8 @@ class Pad(Transform): _v1_transform_cls = _transforms.Pad - _transformed_types = Transform._transformed_types + (is_cvcuda_tensor,) + 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() diff --git a/torchvision/transforms/v2/functional/_geometry.py b/torchvision/transforms/v2/functional/_geometry.py index 2463d130974..2d1d02fd3d7 100644 --- a/torchvision/transforms/v2/functional/_geometry.py +++ b/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, @@ -1682,17 +1683,7 @@ def _pad_with_vector_fill( _pad_image_pil = _register_kernel_internal(pad, PIL.Image.Image)(_FP.pad) -if CVCUDA_AVAILABLE: - _pad_mode_to_cvcuda = { - "constant": cvcuda.Border.CONSTANT, - "reflect": cvcuda.Border.REFLECT101, - "replicate": cvcuda.Border.REPLICATE, - "edge": cvcuda.Border.REPLICATE, - "symmetric": cvcuda.Border.REFLECT, - } - - -def _pad_cvcuda( +def _pad_image_cvcuda( image: "cvcuda.Tensor", padding: list[int], fill: Optional[Union[int, float, list[float]]] = None, @@ -1700,8 +1691,7 @@ def _pad_cvcuda( ) -> "cvcuda.Tensor": cvcuda = _import_cvcuda() - if _pad_mode_to_cvcuda.get(padding_mode) is None: - raise ValueError(f"Padding mode '{padding_mode}' is not supported with CVCUDA") + border_mode = _get_cvcuda_border_from_pad_mode(padding_mode) if fill is None: fill = 0 @@ -1712,7 +1702,7 @@ def _pad_cvcuda( return cvcuda.copymakeborder( image, - border_mode=_pad_mode_to_cvcuda[padding_mode], + border_mode=border_mode, border_value=fill, top=top, left=left, @@ -1722,7 +1712,7 @@ def _pad_cvcuda( if CVCUDA_AVAILABLE: - _register_kernel_internal(pad, _import_cvcuda().Tensor)(_pad_cvcuda) + _register_kernel_internal(pad, _import_cvcuda().Tensor)(_pad_image_cvcuda) @_register_kernel_internal(pad, tv_tensors.Mask) diff --git a/torchvision/transforms/v2/functional/_utils.py b/torchvision/transforms/v2/functional/_utils.py index 11480b30ef9..6f6e99d2f65 100644 --- a/torchvision/transforms/v2/functional/_utils.py +++ b/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 +180,26 @@ 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 _populate_cvcuda_pad_to_border_tables(): + cvcuda = _import_cvcuda() + + global _pad_mode_to_cvcuda_border + + _pad_mode_to_cvcuda_border = { + "constant": cvcuda.Border.CONSTANT, + "reflect": cvcuda.Border.REFLECT101, + "replicate": cvcuda.Border.REPLICATE, + "edge": cvcuda.Border.REPLICATE, + "symmetric": cvcuda.Border.REFLECT, + } + + +def _get_cvcuda_border_from_pad_mode(pad_mode: str) -> "cvcuda.Border": + if len(_pad_mode_to_cvcuda_border) == 0: + _populate_cvcuda_pad_to_border_tables() + return _pad_mode_to_cvcuda_border[pad_mode] From 7d6b6a177795304ad9f8ea3fe8ea79232b8b8bce Mon Sep 17 00:00:00 2001 From: Justin Davis Date: Thu, 4 Dec 2025 13:42:05 -0800 Subject: [PATCH 13/15] remove unneeded cvcuda refs --- torchvision/transforms/v2/functional/_augment.py | 11 +---------- torchvision/transforms/v2/functional/_color.py | 12 +----------- 2 files changed, 2 insertions(+), 21 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: From 53e144f82e9328c46acce78d5b4703f7526251c1 Mon Sep 17 00:00:00 2001 From: Justin Davis Date: Thu, 4 Dec 2025 18:18:29 -0800 Subject: [PATCH 14/15] refactor pad to border mode setup --- .../transforms/v2/functional/_utils.py | 28 ++++++++----------- 1 file changed, 12 insertions(+), 16 deletions(-) diff --git a/torchvision/transforms/v2/functional/_utils.py b/torchvision/transforms/v2/functional/_utils.py index 6f6e99d2f65..91000312489 100644 --- a/torchvision/transforms/v2/functional/_utils.py +++ b/torchvision/transforms/v2/functional/_utils.py @@ -185,21 +185,17 @@ def _is_cvcuda_tensor(inpt: Any) -> bool: _pad_mode_to_cvcuda_border: dict[str, "cvcuda.Border"] = {} -def _populate_cvcuda_pad_to_border_tables(): - cvcuda = _import_cvcuda() - - global _pad_mode_to_cvcuda_border - - _pad_mode_to_cvcuda_border = { - "constant": cvcuda.Border.CONSTANT, - "reflect": cvcuda.Border.REFLECT101, - "replicate": cvcuda.Border.REPLICATE, - "edge": cvcuda.Border.REPLICATE, - "symmetric": cvcuda.Border.REFLECT, - } - - def _get_cvcuda_border_from_pad_mode(pad_mode: str) -> "cvcuda.Border": if len(_pad_mode_to_cvcuda_border) == 0: - _populate_cvcuda_pad_to_border_tables() - return _pad_mode_to_cvcuda_border[pad_mode] + 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 From 5602765f0ec941642a98fb63e579fc3ffd3b9164 Mon Sep 17 00:00:00 2001 From: Justin Davis Date: Fri, 12 Dec 2025 10:43:39 -0800 Subject: [PATCH 15/15] minimize diff --- torchvision/transforms/v2/_geometry.py | 3 +-- torchvision/transforms/v2/_utils.py | 5 ++--- torchvision/transforms/v2/functional/_meta.py | 22 +------------------ torchvision/transforms/v2/functional/_misc.py | 11 ++-------- 4 files changed, 6 insertions(+), 35 deletions(-) diff --git a/torchvision/transforms/v2/_geometry.py b/torchvision/transforms/v2/_geometry.py index ee573a1374e..e9f6edaf043 100644 --- a/torchvision/transforms/v2/_geometry.py +++ b/torchvision/transforms/v2/_geometry.py @@ -465,8 +465,7 @@ class Pad(Transform): _v1_transform_cls = _transforms.Pad - if CVCUDA_AVAILABLE: - _transformed_types = Transform._transformed_types + (_is_cvcuda_tensor,) + _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() diff --git a/torchvision/transforms/v2/_utils.py b/torchvision/transforms/v2/_utils.py index e803aa49c60..bb6051b4e61 100644 --- a/torchvision/transforms/v2/_utils.py +++ b/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, _is_cvcuda_tensor +from torchvision.transforms.v2.functional._utils import _FillType, _FillTypeJIT 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)) } if not chws: raise TypeError("No image or video was found in the sample") @@ -207,7 +207,6 @@ 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/_meta.py b/torchvision/transforms/v2/functional/_meta.py index af03ad018d4..6b8f19f12f4 100644 --- a/torchvision/transforms/v2/functional/_meta.py +++ b/torchvision/transforms/v2/functional/_meta.py @@ -51,16 +51,6 @@ 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) @@ -97,16 +87,6 @@ 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) @@ -145,7 +125,7 @@ def get_size_image_cvcuda(image: "cvcuda.Tensor") -> list[int]: if CVCUDA_AVAILABLE: - _register_kernel_internal(get_size, _import_cvcuda().Tensor)(get_size_image_cvcuda) + _get_size_image_cvcuda = _register_kernel_internal(get_size, cvcuda.Tensor)(get_size_image_cvcuda) @_register_kernel_internal(get_size, tv_tensors.Video, tv_tensor_wrapper=False) diff --git a/torchvision/transforms/v2/functional/_misc.py b/torchvision/transforms/v2/functional/_misc.py index 0fa05a2113c..daf263df046 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 import PIL.Image import torch @@ -13,14 +13,7 @@ from ._meta import _convert_bounding_box_format -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 +from ._utils import _get_kernel, _register_kernel_internal, is_pure_tensor def normalize(