PyTorch Object Detector Training Implementation and Prediction Update

[f4str]

Description

Implementation of the fit method for the PyTorchObjectDetector, PyTorchFasterRCNN, and PyTorchYolo object detectors. This allows these models to be trained using ART. The notebook in notebooks/poisoning_attack_bad_det_rma.ipynb has been updated to demonstrate training these models. This is a partial implementation of #2058 as the fit method needs to also be implemented for the TensorFlow object detectors.

The predict method in the PyTorchObjectDetector and PyTorchYolo classes has been rewritten to implement the following new features:

• Cleanup and refactoring so the code is easier to read and understand.
• The torch.no_grad() scope is used like the PyTorchClassifier to prevent extraneous gradients accumulating in the model which causes slow down.
• Inputs are now batched so the entire input is not processed at once.
• The channels_first parameter is now used to determine whether the input needs to be transformed. The default behavior remains the same.

Additionally, the unit tests for the test_pytorch_object_detector.py and test_pytorch_faster_rcnn.py were rewritten in pytest with a new test added for model training.

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.

• Tests for PyTorchObjectDetector (rewritten in pytest)
• Tests for PyTorchFasterRCNN (rewritten in pytest)
• Tests for PyTorchYolo

Test Configuration:

• OS
• Python version
• ART version or commit number
• TensorFlow / Keras / PyTorch / MXNet version

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 #2067 (f774359) into dev_1.14.0 (f48d73e) will increase coverage by 3.47%.
The diff coverage is 77.33%.

📣 This organization is not using Codecov’s GitHub App Integration. We recommend you install it so Codecov can continue to function properly for your repositories. Learn more

@@              Coverage Diff               @@
##           dev_1.14.0    #2067      +/-   ##
==============================================
+ Coverage       77.23%   80.70%   +3.47%     
==============================================
  Files             294      294              
  Lines           26212    26322     +110     
  Branches         4797     4827      +30     
==============================================
+ Hits            20244    21244    +1000     
+ Misses           4822     3914     -908     
- Partials         1146     1164      +18     

Impacted Files	Coverage Δ
...estimators/object_detection/pytorch_faster_rcnn.py	100.00% <ø> (+22.22%)	⬆️
art/estimators/object_detection/pytorch_yolo.py	75.43% <75.00%> (+60.98%)	⬆️
...mators/object_detection/pytorch_object_detector.py	76.79% <79.72%> (+64.15%)	⬆️

... and 24 files with indirect coverage changes

[beat-buesser]

Hi @f4str Thank you very much for your pull request. I have added a few review comments, what do you think?

[beat-buesser]

Suggested change

	# Apply postprocessing
	predictions = self._apply_postprocessing(preds=results_list, fit=False)
	return predictions # type: ignore
	return predictions

[beat-buesser]

Suggested change

	# Apply postprocessing
	predictions = self._apply_postprocessing(preds=results_list, fit=False)
	return predictions # type: ignore
	return predictions

[beat-buesser]

Suggested change

	# Apply postprocessing
	predictions = self._apply_postprocessing(preds=results_list, fit=False)
	return predictions # type: ignore
	return predictions

[beat-buesser]

@f4str Looks good to me! Thank you very much.