ART 1.7.1

[beat-buesser]

This release of ART 1.7.1 provides updates to ART 1.7.

Added

• Added wrapper Mp3CompressionPyTorch for Mp3Compression to make it compatible with PyTorch-specific attack implementations. (Pytorch wrapper MP3 defense #1210)
• Added new install option non-framework to setup.py to install all non-framework dependencies of ART. (Add non-framework install option #1209)
• Added wrapper VideoCompressionPyTorch for VideoCompression to make it compatible with PyTorch-specific attack implementations. (Pytorch wrapper MP3 defense #1210)

Changed

• Changed Mp3Compression to add back reapplication of normalization to the compressed result. (Pytorch wrapper MP3 defense #1210)
• Changed KerasClassifier.fit to use batching provided by the method fit of the Keras model. (Fix get_started_keras #1182)

Removed

[None]

Fixed

• Fixed bug of not passing user-provided device type, and instead always using default gpu, to standardisation preprocessor in all PyTorchEstimator by using user-provided device type. (Fix device_type of default preprocessor of PyTorchEstimator #1223)
• Fixed bug in method BaseEstimator.fit_generator for fitting generators in cases where preprocessing is defined to not apply preprocessing twice. (Remove duplicate application of preprocessing in fallback BaseEstimator.fit_generator #1219)
• Fixed bug in ImperceptibleASRPyTorch to prevent NaN loss value for batch size larger than 1 by removing unnecessary zero-padding. (ASR attack supports batch size larger than 1 #1198)
• Fixed two bugs in OverTheAirFlickeringPyTorch by making sure that the regularization norms are computed over the whole batch of perturbations, rather than per sample's perturbation and second that the "roll" operations are performed over the batch samples, rather than over the frames. (norms per sample; and roll left, right over frames, not batch #1192)
• Fixed bug in SpectralSignatureDefense, that lead to rejections of all clean images, by correctly indexing the label data. (Fix bugs in Spectral Signature Defense #1189)
• Fixed bug of accidentally removed checks for apply_fit and apply_predict properties of framework-independent Preprocessor tools in PyTorchEstimator and TensorFlowV2Estimator. With the bug the Preprocessor tools were always applied in methods fit and predict independent of the values of apply_fit and apply_predict. (Add back apply_fit and apply_predict check for Numpy-based preprocessing #1181)
• Fixed bug in MembershipInferenceBlackBoxRemove.infer by removing unnecessary shuffling of the test data. (Fix shuffling of test data in infer (#1165) #1173)
• Fixed bug in PixelAttack and ThresholdAttack by casting input data to correct dtype. (Minor bug of Dtype casting in Pixel and Threshold Attack #1175)

---

This discussion was created from the release ART 1.7.1.