TensorFlow Addons is end of life in May 2024

Trusted-AI / adversarial-robustness-toolbox

Adversarial Robustness Toolbox (ART) - Python Library for Machine Learning Security - Evasion, Poisoning, Extraction, Inference - Red and Blue Teams

https://adversarial-robustness-toolbox.readthedocs.io/en/latest/

MIT License

4.76k stars 1.15k forks source link

TensorFlow Addons is end of life in May 2024 #2316

Open jamesmyatt opened 11 months ago

jamesmyatt commented 11 months ago

TensorFlow Addons (TFA) has ended development and introduction of new features. TFA has entered a minimal maintenance and release mode until a planned end of life in May 2024. Please modify downstream libraries to take dependencies from other repositories in our TensorFlow community (e.g. Keras, Keras-CV, and Keras-NLP).

For more information see: https://github.com/tensorflow/addons/issues/2807

What are the plans for removing/replacing TFA?

jamesmyatt commented 10 months ago

I had a little look at this and I thought that the easiest solution was probably to vendor the functions from tfa.image that are used in this framework. What do you think?

I think they are (based on https://github.com/search?q=repo%3ATrusted-AI%2Fadversarial-robustness-toolbox+tfa.&type=code):

tfa.image.random_cutout
tfa.image.median_filter2d
tfa.image.transform
tfa.image.rotate

beat-buesser commented 10 months ago

Hi @jamesmyatt Thank you for the update, we have noticed it too. I'm currently working on updating ART and it's testing for Tensorflow 2.14. I have not yet found replacement function for all functions we use from TFA. Would you be interested to take a closer look at ART's dependencies and how to best replace them?

jamesmyatt commented 10 months ago

I think that:

tfa.image.transform -> Keras has a function that appears to match (https://github.com/keras-team/keras/blob/v2.14.0/keras/layers/preprocessing/image_preprocessing.py#L720)
tfa.image.rotate -> can be implemented via the transform function above (c.f. https://github.com/keras-team/keras/blob/68f9af408a1734704746f7e6fa9cfede0d6879d8/keras/layers/preprocessing/image_preprocessing.py#L987). Although the layer could probably be refactored to extract the random rotation (https://keras.io/api/layers/preprocessing_layers/image_augmentation/random_rotation/)
tfa.image.random_cutout -> Keras CV (https://keras.io/api/keras_cv/layers/preprocessing/random_cutout/)
tfa.image.median_filter2d -> probably just copy-and-paste this or replace it with a different filter from tfio.experimental.filter