ToDtype CV-CUDA Backend #9278

justincdavis · 2025-11-19T21:44:54Z

Summary

Add the backend kernel for ToDtype transform using CV-CUDA

How to use

import cvcuda
import torchvision.transforms.v2.functional as F

cvc_tensor = cvcuda.Tensor((1, 224, 224, 3), cvcuda.Type.U8, cvcuda.TensorLayout.NHWC)
# Dispatches to F.to_dtype_cvcuda
cvc_fp32_tensor = F.to_dtype(cvc_tensor, torch.float32)

pytorch-bot · 2025-11-19T21:44:58Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/vision/9278

📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

This comment was automatically generated by Dr. CI and updates every 15 minutes.

meta-cla · 2025-11-19T21:45:00Z

Hi @justincdavis!

Thank you for your pull request and welcome to our community.

Action Required

In order to merge any pull request (code, docs, etc.), we require contributors to sign our Contributor License Agreement, and we don't seem to have one on file for you.

Process

In order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA.

Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with CLA signed. The tagging process may take up to 1 hour after signing. Please give it that time before contacting us about it.

If you have received this in error or have any questions, please contact us at cla@meta.com. Thanks!

AntoineSimoulin

Hey @justincdavis, thanks a lot for the PR. I left some comments and questions as a first review. Let me know what you think!

test/test_transforms_v2.py

torchvision/transforms/v2/functional/_misc.py

… setup

AntoineSimoulin · 2025-11-26T14:55:54Z

@justincdavis could you complete the missing Contributor License Agreement (c.f. earlier comment from the meta-cla bot)?

AntoineSimoulin

Hey @justincdavis, thanks for addressing my first round of comments. I had another pass. Will it be possible to have another iteration on the PR based on my new comments? Thanks a lot for your time here!

torchvision/transforms/v2/functional/_misc.py

test/test_transforms_v2.py

zy1git

Hi,

This is just a light pass of the review. Let me know what you think.

test/common_utils.py

test/test_transforms_v2.py

torchvision/transforms/v2/functional/_meta.py

justincdavis · 2025-12-04T18:55:21Z

Hi @zy1git thanks for the first pass! I have updated this PR to reflect the conventions of the flip PR, LMK what you think!

…sform class

NicolasHug

Thanks a lot for the PR @justincdavis , I left a first pass

NicolasHug · 2025-12-09T17:20:17Z

test/test_transforms_v2.py

-                make_image_cvcuda,
-                marks=pytest.mark.skipif(not CVCUDA_AVAILABLE, reason="CVCUDA is not available"),
-            ),
+            pytest.param(make_image_cvcuda, marks=CV_CUDA_TEST),


Just a not that you should be able to remove these changes once #9305 lands.

NicolasHug · 2025-12-09T17:28:41Z

torchvision/transforms/v2/functional/_utils.py

+
+
+def _get_cvcuda_type_from_torch_dtype(dtype: torch.dtype) -> "cvcuda.Type":
+    if len(_torch_to_cvcuda_dtypes) == 0:


Suggested change

if len(_torch_to_cvcuda_dtypes) == 0:

if not _torch_to_cvcuda_dtypes:

NicolasHug · 2025-12-09T17:29:11Z

torchvision/transforms/v2/functional/_utils.py

+
+
+def _get_torch_dtype_from_cvcuda_type(dtype: "cvcuda.Type") -> torch.dtype:
+    if len(_cvcuda_to_torch_dtypes) == 0:


Suggested change

if len(_cvcuda_to_torch_dtypes) == 0:

if not _cvcuda_to_torch_dtypes):

NicolasHug · 2025-12-10T14:32:21Z

test/test_transforms_v2.py

+    def test_functional_signature(self, kernel, input_type):
+        if kernel is F._misc._to_dtype_image_cvcuda:
+            input_type = _import_cvcuda().Tensor
+        check_functional_kernel_signature_match(F.to_dtype, kernel=kernel, input_type=input_type)


Thanks for adding this test!

NicolasHug · 2025-12-10T14:33:52Z

torchvision/transforms/v2/functional/_meta.py

    return get_dimensions_image(video)


+def get_dimensions_image_cvcuda(image: "cvcuda.Tensor") -> list[int]:


QQ are these changes needed here in this PR? Same q for get_num_channels_image_cvcuda

NicolasHug · 2025-12-10T14:35:28Z

torchvision/transforms/v2/functional/_misc.py

+if TYPE_CHECKING:
+    import cvcuda  # type: ignore[import-not-found]
+if CVCUDA_AVAILABLE:
+    cvcuda = _import_cvcuda()  # noqa: F811


I think we'll want to always use _import_cvcuda() instead of defining the global cvcuda module here, it's safer. Having

cvcuda = _import_cvcuda()

within a function like you did in _to_dtype_image_cvcuda is probably OK though.

NicolasHug · 2025-12-10T14:38:33Z

torchvision/transforms/v2/functional/_utils.py

    except ImportError:
        return False
+
+


All the stuff below: unless we already know we'll need it elsewhere, I'd suggested to just implement that within _misc.py instead of in _utils.py. Perhaps you needed the same functionality in another transform that's not in _misc.py? In which case, it's OK to have it here.

NicolasHug · 2025-12-10T14:39:03Z

torchvision/transforms/v2/_misc.py

+    if CVCUDA_AVAILABLE:
+        _transformed_types = Transform._transformed_types + (_is_cvcuda_tensor,)


I think we don't need to protect that with CVCUDA_AVAILABLE anymore

NicolasHug · 2025-12-10T15:05:11Z

test/test_transforms_v2.py

+        if is_uint16_to_uint8:
+            atol = 255
+        elif is_uint8_to_uint16 and not scale:
+            atol = 255


IIUC, this 255 tol is needed because in torch, when scale is False, we're doing a brutal .to(dtype) which is going to cause a lot of overflows, whereas in CVCUDA you either cap the result or always scale?

I'm hoping we can simplify this a bit, potentially by dropping support for uint8 <-> uint16 conversions when scale is False on CV-CUDA. I feel like that's not a really valid conversion to support anyway. The general idea is that for all transforms, we'll want the CVCUDA backend to have very close results to the existing tensor backend. A difference of 255 is too large.

BTW, we should be able to set atol to 0 or 1 when is_uint16_to_uint8 and scale is True?

AntoineSimoulin reviewed Nov 24, 2025

View reviewed changes

implement additional cvcuda infra for all branches to avoid duplicate…

44db71c

… setup

AntoineSimoulin mentioned this pull request Nov 25, 2025

Implement Flip transforms with CVCUDA backend #9277

Merged

update make_image_cvcuda to have default batch dim

e3dd700

justincdavis mentioned this pull request Nov 26, 2025

GaussianBlur CV-CUDA Backend #9280

Open

AntoineSimoulin reviewed Nov 26, 2025

View reviewed changes

torchvision/transforms/v2/functional/_misc.py Outdated Show resolved Hide resolved

torchvision/transforms/v2/functional/_misc.py Outdated Show resolved Hide resolved

test/test_transforms_v2.py Outdated Show resolved Hide resolved

meta-cla bot added the cla signed label Dec 2, 2025

justincdavis added 4 commits December 1, 2025 18:16

add stanardized setup to main for easier updating of PRs and branches

c035df1

update is_cvcuda_tensor

98d7dfb

add cvcuda to pil compatible to transforms by default

ddc116d

remove cvcuda from transform class

e51dc7e

zy1git reviewed Dec 4, 2025

View reviewed changes

test/common_utils.py Show resolved Hide resolved

test/test_transforms_v2.py Outdated Show resolved Hide resolved

torchvision/transforms/v2/functional/_meta.py Outdated Show resolved Hide resolved

merge with main

e14e210

justincdavis force-pushed the feat/dtype_cvcuda branch from 204f698 to 4259d7f Compare December 4, 2025 18:52

justincdavis added 12 commits December 4, 2025 11:07

resolve more formatting naming

4939355

initial draft of to_dtype_cvcuda

ec76196

fix: to_dtype_cvcuda conventions

bd823cf

remove staticmethod from reference todtype

f7aa94a

add docstring for explain scaling setup, combine correctness checks

b21d9f0

resolve more review comments

973e058

simplify todtype testing

d871331

add int -> int scaling setup for cvcuda, use bit diff for scale

736a2e6

further simplify todtype test

7a231b1

update todtype based on PR reviews

d3e4573

cleanup commnet, variable names

ec93ba3

update to_dtype_cvcuda name

89122db

justincdavis added 3 commits December 4, 2025 11:07

update to standards from flip PR

1b0d295

remove cvcuda updates to augment

009f925

remove cvcuda refs from color

41af724

justincdavis force-pushed the feat/dtype_cvcuda branch from 7e17ce4 to 41af724 Compare December 4, 2025 19:07

justincdavis and others added 12 commits December 4, 2025 13:18

refactor dtype converters to be in utils

d12e4df

add type checking for cvcuda

c198cf0

provide better error for todtype

18df67f

refactor to simplify setup for dtype conversions

c5a2a5a

Merge branch 'main' into feat/dtype_cvcuda

915ffb1

fix: not testing transform class correctness in ToDtype, resolved

7f41c95

preserve previous torchvision test behavior for non cvcuda inputs

9b41552

further simplify branching flow of testtodtype image correctness

b9c378b

add functional signature tests, fix bug in type check in todtype tran…

1781244

…sform class

add consolidated cvcuda test markers

626b47a

finalize consolidated cvcuda skip behavior

e8540ba

revert var name change back to input

5aa4b3d

NicolasHug reviewed Dec 10, 2025

View reviewed changes



		def _get_cvcuda_type_from_torch_dtype(dtype: torch.dtype) -> "cvcuda.Type":
		if len(_torch_to_cvcuda_dtypes) == 0:

	if len(_torch_to_cvcuda_dtypes) == 0:
	if not _torch_to_cvcuda_dtypes:



		def _get_torch_dtype_from_cvcuda_type(dtype: "cvcuda.Type") -> torch.dtype:
		if len(_cvcuda_to_torch_dtypes) == 0:

	if len(_cvcuda_to_torch_dtypes) == 0:
	if not _cvcuda_to_torch_dtypes):

		return get_dimensions_image(video)


		def get_dimensions_image_cvcuda(image: "cvcuda.Tensor") -> list[int]:

		if CVCUDA_AVAILABLE:
		_transformed_types = Transform._transformed_types + (_is_cvcuda_tensor,)

ToDtype CV-CUDA Backend #9278

Are you sure you want to change the base?

ToDtype CV-CUDA Backend #9278

Uh oh!

Conversation

justincdavis commented Nov 19, 2025

Summary

How to use

Uh oh!

pytorch-bot bot commented Nov 19, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/vision/9278

Uh oh!

meta-cla bot commented Nov 19, 2025

Action Required

Process

Uh oh!

AntoineSimoulin left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

AntoineSimoulin commented Nov 26, 2025

Uh oh!

AntoineSimoulin left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

zy1git left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

justincdavis commented Dec 4, 2025

Uh oh!

NicolasHug left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

pytorch-bot bot commented Nov 19, 2025 •

edited

Loading