Background

• Describe the status of the problem you wish to solve
• Attach the relevant issue if have

Introduction

• In the attack algorithm, there is DiverseInputIterativeMethod (DI2-FGSM). This method uses data transformation to increase the transferability of adversarial examples. Currently, we use random sizing, padding, and adding noise to the input to generate adversarial examples. We provide other data transformation methods in utils/image_transform.py, please use other transfom combination methods to generate adversarial samples.

Trail