Adversarial example classified correctly
mkbera opened this issue · 1 comments
mkbera commented
The adversarial samples like 'occl_4_to_7_model1.png' are being classifed correctly by model1.
peikexin9 commented
The images generated undergo further processing only for illustration purpose so I believe they may not trigger misclassification.
I would recommend saving the raw numpy images matrices and reload them to check if they can really misguide the prediction.