This repository is the official Pytorch code implementation for our paper Leveraging Spatial Invariance to Boost Adversarial Transferability.

Requirements

-python 3.6
-torch 1.8.1
-pretrainedmodels 0.7.4
-numpy 1.19.5
-Pillow 8.4.0
-pandas 1.2

Implementation

Models

Download pretrained PyTorch models here. Then put these models into ./models/.

Datase

Please refer to SSA. Then put dataset into ./dataset/.

Generate adversarial examples

Running attack.py to generate adversarial examples.

Evaluations

Running verify.py to evaluate the attack success rate.