Implement estimator for Hugging Face models

[GiulioZizzo]

Description

In this PR we introduce an estimator to support HF models for classification. This can enable initial support for HF in ART which we can expand on and develop in future.

Fixes #2225

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

Please describe the tests that you ran to verify your changes. Consider listing any relevant details of your test configuration.

• Inclusion of a HF workflow to run the test suite with a HF classifier. Updates to core tests to explicitly check expected HF performance.

Test Configuration:

• OS: Mac/RHEL
• Python version: 3.8/3.11
• ART version or commit number: 1.15.1
• TensorFlow / Keras / PyTorch / MXNet version: TF 2.12, Pytorch 2.0.1, keras 2.12

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

[codecov-commenter]

Codecov Report

Merging #2245 (6a6f094) into dev_1.16.0 (cbbd91c) will increase coverage by 0.95%.
The diff coverage is 89.23%.

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

@@              Coverage Diff               @@
##           dev_1.16.0    #2245      +/-   ##
==============================================
+ Coverage       84.68%   85.64%   +0.95%     
==============================================
  Files             323      324       +1     
  Lines           29176    29304     +128     
  Branches         5374     5400      +26     
==============================================
+ Hits            24709    25098     +389     
+ Misses           3115     2828     -287     
- Partials         1352     1378      +26     

Files Changed	Coverage Δ
...cks/poisoning/perturbations/image_perturbations.py	87.64% <53.84%> (-5.95%)	⬇️
art/estimators/classification/hugging_face.py	92.98% <92.98%> (ø)
...rigger_backdoor/hidden_trigger_backdoor_pytorch.py	75.42% <100.00%> (+2.14%)	⬆️
art/estimators/classification/__init__.py	100.00% <100.00%> (ø)

... and 8 files with indirect coverage changes

[github-advanced-security]

CodeQL found more than 10 potential problems in the proposed changes. Check the Files changed tab for more details.

[beat-buesser]

Hi @GiulioZizzo Thank you very much for your pull request. I have completed my first review, please let me know what you think?

[beat-buesser]

Please remove the large model and integrate get_image_classifier_hf into get_mnist_classifier.

[GiulioZizzo]

Could we call image_dl_estimator instead?

[beat-buesser]

Hi @GiulioZizzo I have added a few suggestions. Please let me know what you think!

[beat-buesser]

Does the transpose with 4 indices work if len(shape)==3?

[GiulioZizzo]

Good point: that should only apply to 4D inputs and an additional one is needed for 2D to go from HC to CH.