Skip to content

Latest commit

 

History

History

Adv

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
models
models
 
 
scripts
scripts
 
 
README.md
README.md
 
 
attack_algo.py
attack_algo.py
 
 
functions.py
functions.py
 
 
requirements.txt
requirements.txt
 
 
test.py
test.py
 
 
train.py
train.py
 
 

README.md

Adversarial Training with SaBN

Environment

  • Ubuntu 16.04
  • NVIDIA GPU
  • python >= 3.6

Quick start

Installation:

  1. Install pytorch >= v1.1.0 following official instruction.
  2. Clone this repo:
git clone https://github.com/VITA-Group/Sandwich-Batch-Normalization
cd Adv
  1. Install dependencies:
pip install -r requirements.txt

Usage

Training

bash scripts/train_bn.sh
bash scripts/train_auxbn.sh
bash scripts/train_saauxbn.sh

Testing

Check Tensorboard:

tensorboard --logdir output --port 6001

Results

The evaluation results:

Evaluation BN AuxBN (clean branch) SaAuxBN (clean branch) (ours)
Clean (SA) 84.84 94.47 94.62
Evaluation BN AuxBN (adv branch) SaAuxBN (adv branch) (ours)
Clean (SA) 84.84 83.42 84.08
PGD-10 (RA) 41.57 43.05 44.93
PGD-20 (RA) 40.02 41.60 43.14

The visualization of test loss:

Test loss Adversarial test loss
CIFAR100 ImageNet