RYC-98 / FSD-MIM-and-NPGA

Official codes for FSD-MIM (Accepted by Applied Soft Computing)

Frequency-based methods for improving the imperceptibility and transferability of adversarial examples

Requirements

• python 3.8
• torch 1.8
• numpy 1.19
• pandas 1.2

Implementation

• Generate adversarial examples

  Using FSD_MIM.py to generate highly transferable adversarial examples, you can run this attack as following

  CUDA_VISIBLE_DEVICES=gpuid python FSD_MIM.py --output_dir outputs

  where gpuid can be set to any free GPU ID in your machine. And adversarial examples will be generated in directory ./outputs.

• Evaluations on normally trained models

  Running verify.py to evaluate the attack success rate

  CUDA_VISIBLE_DEVICES=gpuid python verify.py

Main Results

Number of augmented copies is set to 5 for each iteration.