Total Variance Minimization with PyTorch support

[MatteoFasulo]

Description

Implemented Total Variation Minimization using PyTorch with GPU support, closely modeled after the existing CPU-based scipy.optimize.minimize approach. The CPU version iteratively minimizes over each image in the batch and each channel individually, resulting in time complexity of O(B × C) where B is the batch size and C the number of channels. Additionally, loss computation on the CPU introduces extra latency per iteration. The new PyTorch-based implementation enables parallelized, accelerated computation directly on the GPU, significantly improving runtime performance.

Type of change

Please check all relevant options.

• Improvement (non-breaking)
• Bug fix (non-breaking)
• New feature (non-breaking)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• This change requires a documentation update

Testing

Extensive test coverage was added via

pytest tests/defences/preprocessor/test_variance_minimization_pytorch.py --framework="pytorch"

• test_variance_minimization_pytorch.py

validating:

• Memory efficiency
• Input dimension and shape handling
• Channel ordering (first vs. last)
• Output clipping within expected value ranges
• Convergence behavior
• Reproducibility across multiple runs with consistent seeds

Test Configuration:

• OS: Ubuntu 24.04 LTS
• Python version: Python 3.10.15
• ART version or commit number: 1.19.2
• TensorFlow / Keras / PyTorch / MXNet version: PyTorch

Checklist

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• My changes have been tested using both CPU and GPU devices

[beat-buesser]

Hi @MatteoFasulo Thank you for your pull request, I'll add a review in the next few days.

[codecov]

Codecov Report

❌ Patch coverage is 92.13483% with 7 lines in your changes missing coverage. Please review.
✅ Project coverage is 85.18%. Comparing base (df6b772) to head (8449450).
⚠️ Report is 9 commits behind head on dev_1.21.0.

Files with missing lines	Patch %	Lines
...nces/preprocessor/variance_minimization_pytorch.py	92.13%	4 Missing and 3 partials ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           dev_1.21.0    #2673   +/-   ##
===========================================
  Coverage       85.17%   85.18%           
===========================================
  Files             329      330    +1     
  Lines           29731    29820   +89     
  Branches         4991     5007   +16     
===========================================
+ Hits            25324    25402   +78     
- Misses           2976     2983    +7     
- Partials         1431     1435    +4     

Files with missing lines	Coverage Δ
...nces/preprocessor/variance_minimization_pytorch.py	92.13% <92.13%> (ø)

... and 4 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[beat-buesser]

@MatteoFasulo Could you please fix the style checks and CodeQL alerts above? Run Black formatter (version black==25.1.0) with black --line-length 120 art/ and black --line-length 120 tests/ to fix the formatting.

[MatteoFasulo]

@MatteoFasulo Could you please fix the style checks and CodeQL alerts above? Run Black formatter (version black==25.1.0) with black --line-length 120 art/ and black --line-length 120 tests/ to fix the formatting.

Sure, I will do that in the next few days.

[beat-buesser]

Hi @MatteoFasulo Thank you very much for your pull request. I have added 3 minor change requests, could you please push the changes?

[MatteoFasulo]

Hi @MatteoFasulo Thank you very much for your pull request. I have added 3 minor change requests, could you please push the changes?

Sure, I added the three suggested changes in the last commit. I removed the unused numpy import, added the eps parameter type hint and documentation.

[beat-buesser]

Suggested change

	if TYPE_CHECKING:
	from art.utils import CLIP_VALUES_TYPE
	if TYPE_CHECKING:
	import torch
	from art.utils import CLIP_VALUES_TYPE

[beat-buesser]

Hi @MatteoFasulo Thank you! I have missed this before, an additional import of torch is necessary for the typing to pass the Style Checks. Please add the code suggestion above and run Black formatter again to ensure correct line formatting. Afterwards we should be good to merge.

[MatteoFasulo]

Hi @MatteoFasulo Thank you! I have missed this before, an additional import of torch is necessary for the typing to pass the Style Checks. Please add the code suggestion above and run Black formatter again to ensure correct line formatting. Afterwards we should be good to merge.

Hi @beat-buesser,
I added the missing import of torch and checked with black formatter. It should be good right now. Feel free to reach out if there is something wrong.

[beat-buesser]

Hi @MatteoFasulo Thank you very much for your contribution to ART!