GaussianBlur CV-CUDA Backend #9280

justincdavis · 2025-11-19T21:51:26Z

Summary

Implement the CV-CUDA backend kernel for gaussian_blur

How to use

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

cv_tensor = cvcuda.Tensor((1, 224, 224, 3), cvcuda.Type.U8, cvcuda.TensorLayout.NHWC)
# dispatched to F.gaussian_blur_cvcuda
blurred = F.gaussian_blur(cv_tensor, (5, 5))

pytorch-bot · 2025-11-19T21:51:30Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/vision/9280

📄 Preview Python docs built from this PR

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

✅ No Failures

As of commit 754223f with merge base aa35ca1 ():
💚 Looks good so far! There are no failures yet. 💚

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

meta-cla · 2025-11-19T21:51:32Z

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

Thanks for the PR @justincdavis. Left a few comments, looking good otherwise!

torchvision/transforms/v2/functional/_misc.py

test/test_transforms_v2.py

AntoineSimoulin · 2025-11-24T15:18:12Z

test/test_transforms_v2.py

+        actual_torch = F.cvcuda_to_tensor(actual)
+
+        if dtype.is_floating_point:
+            torch.testing.assert_close(actual_torch, expected, rtol=0, atol=0.3)


Why setting atol=0.3 here?

Good question! I added a comment on atol=0.3, most likely from floating point differences between the underlying filter2d in CV-CUDA compared to torch.conv2d. Let me know if you want more explanation and/or something else here.

Should we set it as in test_functional_image_correctness with torch.testing.assert_close(actual, expected, rtol=0, atol=1) for consistency?

@AntoineSimoulin I ended up rewriting the test setup, and moved all the tests into a single block. Both CV-CUDA and torchvision share the same assert statement now. LMK if you think it looks like a good change.

torchvision/transforms/v2/functional/_misc.py

… setup

meta-cla · 2025-12-02T01:21:41Z

Thank you for signing our Contributor License Agreement. We can now accept your code for this (and any) Meta Open Source project. Thanks!

…import, explicit imports in func

…igma

zy1git

I’ve left some comments on this PR.
Please feel free to address them or reach out if you’d like to discuss any points further.

zy1git · 2025-12-18T01:17:03Z

test/test_transforms_v2.py

        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")


I feel that this comment is bit confusing:

Does it mean "non-(float32 or uint8)" → neither float32 nor uint8?

Or "(non-float32) or uint8" → something else entirely?

Thus, I recommend to use "CVCUDA only supports float32 and uint8 dtypes for gaussian blur".

zy1git · 2025-12-18T01:53:36Z

test/test_transforms_v2.py

+
+        if input_type == "cvcuda.Tensor":
+            actual = F.cvcuda_to_tensor(actual)
+            actual = actual.squeeze(0).to(device=device)


We can also use actual=actual[0].to(device=device) since batch size is guaranteed to be 1 in this case. Not sure we need to be consistent to the implementation here: https://github.com/pytorch/vision/pull/9277/changes#diff-9c2dde92db86c123fee225e39b7c1ef96e08a3e79a9dcc9a2d68b21ed51a81d0R1315

zy1git · 2025-12-18T01:55:50Z

test/test_transforms_v2.py

+
+        if input_type == "cvcuda.Tensor":
+            actual = F.cvcuda_to_tensor(actual)
+            actual = actual.squeeze(0).to(device=device)


I think we can also use actual = actual[0].to(device=device) since the batch size is always 1 in this case. Not sure we need to be consistent with the implementation here: https://github.com/pytorch/vision/pull/9277/changes#diff-9c2dde92db86c123fee225e39b7c1ef96e08a3e79a9dcc9a2d68b21ed51a81d0R1315

zy1git · 2025-12-18T09:50:35Z

test/test_transforms_v2.py

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


See this PR: https://github.com/pytorch/vision/pull/9305/changes

There are other parts with similar issues that also need to be addressed.

AntoineSimoulin reviewed Nov 24, 2025

View reviewed changes

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

justincdavis added 2 commits November 25, 2025 09:14

implement additional cvcuda infra for all branches to avoid duplicate…

44db71c

… setup

update make_image_cvcuda to have default batch dim

e3dd700

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

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

c035df1

justincdavis force-pushed the feat/gaussian_cvcuda branch from 75e4b20 to 1cb4629 Compare December 2, 2025 02:22

update is_cvcuda_tensor

98d7dfb

justincdavis force-pushed the feat/gaussian_cvcuda branch from 1cb4629 to 2c6c99f Compare December 2, 2025 02:25

justincdavis added 14 commits December 2, 2025 12:37

add cvcuda to pil compatible to transforms by default

ddc116d

remove cvcuda from transform class

e51dc7e

merge with main

e14e210

resolve more formatting naming

4939355

draft initial gaussian_blur cvcuda kernel implementation

ac82cea

fix: move cvcuda tests to centralized class, more guards againist no …

b18fedf

…import, explicit imports in func

consolidate gaussian_blur_image to use new validate_kernel_size_and_s…

5df3a7d

…igma

resolve more review comments

9cd7582

match type hint in validate to gaussian_blur_image

d9e3f83

resolve tests failing due to cvcuda.Tensor

ccd4c1d

fix guassian border mode to adhere to torch/opencv

b3d7814

remove unneeded cvcuda setup

160e047

use assert_close

5ce83b1

update gaussian blur with main standards

3bcc517

justincdavis force-pushed the feat/gaussian_cvcuda branch from 0019e9f to 3bcc517 Compare December 4, 2025 19:15

justincdavis and others added 3 commits December 4, 2025 13:44

check input type on kernel for signature test

2edfdff

Merge branch 'main' into feat/gaussian_cvcuda

1155607

minimize diff

8c2cb57

justincdavis added 2 commits December 12, 2025 09:21

drop more unused duplication for shared infra branch

e886fc0

resolve formatting change

71ea23c

zy1git reviewed Dec 18, 2025

View reviewed changes

justincdavis added 2 commits December 18, 2025 14:50

Merge remote-tracking branch 'upstream/main' into feat/gaussian_cvcuda

3681da6

update tests based on comments

754223f

GaussianBlur CV-CUDA Backend #9280

Are you sure you want to change the base?

GaussianBlur CV-CUDA Backend #9280

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/9280

✅ No Failures

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!

AntoineSimoulin Nov 24, 2025

Choose a reason for hiding this comment

Uh oh!

justincdavis Nov 25, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

AntoineSimoulin Nov 26, 2025

Choose a reason for hiding this comment

Uh oh!

justincdavis Nov 26, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

meta-cla bot commented Dec 2, 2025

Uh oh!

zy1git left a comment

Choose a reason for hiding this comment

Uh oh!

zy1git Dec 18, 2025

Choose a reason for hiding this comment

Uh oh!

zy1git Dec 18, 2025

Choose a reason for hiding this comment

Uh oh!

zy1git Dec 18, 2025

Choose a reason for hiding this comment

Uh oh!

zy1git Dec 18, 2025

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

pytorch-bot bot commented Nov 19, 2025 •

edited

Loading

justincdavis Nov 25, 2025 •

edited

Loading