Ask HN: ML Comparing Adversarial ML libaries - Foolbox & IBM ART

Hi HN, We are doing a project to compare Foolbox and IBM's adversarial robustness toolbox. What we are planning is to compare both the libraries implementation of attacks common in both. So we take attack X and see how IBM and Foolbox compares. We are looking at two metrics: 1. Time taken by each library to generate adversarials for the same attack 2. Accuracy of the original model on the adversarials Are there any other metrics that are interesting for this?