Codecov Report

❌ Patch coverage is 98.64865% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 85.22%. Comparing base (293bd22) to head (8e7421b).
⚠️ Report is 31 commits behind head on dev_1.21.0.

@@              Coverage Diff               @@
##           dev_1.21.0    #2720      +/-   ##
==============================================
+ Coverage       83.30%   85.22%   +1.92%     
==============================================
  Files             330      331       +1     
  Lines           29539    29894     +355     
  Branches         5007     5023      +16     
==============================================
+ Hits            24607    25477     +870     
+ Misses           3516     2981     -535     
- Partials         1416     1436      +20     

Update:

1. Worked on the Codecov patch coverage issue (previously 80.82192% with 14 lines missing) by extending unit tests to hit the remaining branches (shape routing for NHWC/NCHW, one‑hot label handling, empty‑class skip, and best‑coord guard).
2. Ran Black and fixed pycodestyle/mypy findings to address the Style Check.

Kindly re‑run CI and re‑review, happy to adjust further if needed. Thank you @beat-buesser

Hi @beat-buesser may I know why there is still 1 pending check regarding PyTorch 2.6.0 (Python 3.10) (Expected — Waiting for status to be reported) ?

I think from the last check, the only issue was the CI Style Checks but now it is successful. The Codecov part also passed this time with higher percentage of patch coverage.

Please let me know if there is something I need to further adjust, thank you!

Hi @beat-buesser may I know why there is still 1 pending check regarding PyTorch 2.6.0 (Python 3.10) (Expected — Waiting for status to be reported) ?

It got replaced by PyTorch 2.8.0, I have now updated the settings for this target branch.

I'll add a review in the coming days.

Hi @nicholasadriel Thank you for your pull request. I have added a few review comments, could you please take a look?

Are you able to reproduce the results of paper by Wu et al. with this code?

Hi @beat-buesser I have reviewed the comments, performed a quick test, and everything's fine.

Regarding reproducing the results of paper by Wu et al., I wasn’t able to reproduce the exact numbers, but the core OPS algorithm here mirrors the reference implementation (single per-class pixel chosen by a stability/deviation criterion; same coordinate and color applied to all images of that class; labels unchanged; model-free).

Reproduction snapshot (CIFAR-10, ResNet-18, 200 epochs):
Wu et al. (paper): Clean accuracy 94.01%, OPS-poisoned accuracy 15.56%, Drop 78.45%
This ART implementation: Clean accuracy 89.22%, OPS-poisoned accuracy 7.64%, Drop 81.58%

While the absolute accuracies differ, the effect size of the attack is very close, which supports that the implementation captures the intended behavior. The gap in absolute numbers is likely due to training-pipeline differences (data preprocessing/augmentation, normalization, optimizer/schedule, weight decay, and seeds).

@nicholasadriel I forgot, we still need from __future__ import annotations for the new typing in some of the test runs.

@nicholasadriel I forgot, we still need from __future__ import annotations for the new typing in some of the test runs.

Hi @beat-buesser thank you for the review, I have added the import statement and corrected few typing issues based on last review results. Please kindly re‑run CI and re‑review, happy to adjust further if needed. Thank you.

@nicholasadriel I forgot, we still need from __future__ import annotations for the new typing in some of the test runs.

Hi @beat-buesser sorry, can you please kindly re‑run CI and re‑review again? I misplaced the from future import annotations thus failed some tests, but I have corrected it just now. Thanks!

Hi @beat-buesser since the last checks have already passed, what will be the next step regarding this pull request? Thanks!

Hi @nicholasadriel I think from a review point of view it is now ready for merging.