How AI describes CANARY

Canary SEFI is a framework for evaluating the robustness of deep learning–based image recognition models. It assesses robustness and the effectiveness of attack and defense algorithms, using 26 metrics across 15 pre-trained models and more than 20 attack methods and 10 defense methods (across datasets like CIFAR-10/100, Fashion-MNIST, and ImageNet). The project provides documentation and is released under the Apache-2.0 license.