Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/gralliry/adversarial-attack-generation-techniques
Method of Generating Adversarial Examples By Pytorch in CIFAR10: L-BFGS, FGSM, I-FGSM, MI-FGSM, DeepFool, C&W, JSMA, ONE-PIXEL, UPSET
https://github.com/gralliry/adversarial-attack-generation-techniques
deep-learning gan neural-network pytorch
Last synced: 3 months ago
JSON representation
Method of Generating Adversarial Examples By Pytorch in CIFAR10: L-BFGS, FGSM, I-FGSM, MI-FGSM, DeepFool, C&W, JSMA, ONE-PIXEL, UPSET
- Host: GitHub
- URL: https://github.com/gralliry/adversarial-attack-generation-techniques
- Owner: gralliry
- Created: 2024-01-12T05:16:27.000Z (about 1 year ago)
- Default Branch: master
- Last Pushed: 2024-08-18T09:51:21.000Z (6 months ago)
- Last Synced: 2024-08-18T11:14:15.380Z (6 months ago)
- Topics: deep-learning, gan, neural-network, pytorch
- Language: Python
- Homepage:
- Size: 1.25 MB
- Stars: 4
- Watchers: 2
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Adversarial Attack Generation Techniques
## File Overview
```
│
├── attack # Folder containing attack model classes
│ └── fgsm.py
│ └── ...
│
├── datasets # Folder for datasets
│ └── download.py # Script to download datasets
│
├── parameter # Folder for model parameters
│ └── ResNet
│ │ └── ...pth
│ └── ...
│ │
│ └── UPSET # Folder for perturbation generation model parameters used by UPSET
│ └── target_0 # Target folders corresponding to perturbation generation model parameters
│ │ └── ...pth
│ └── target_1
│ └── ...
│
├── models
│ └── resnet.py # Recognition model class
│ └── ...
│
├── report # Folder for attack effect visualization
│
├── tensorboard # Folder for tensorboard training process logs
│
├── contrast.py # Visualization of attack effects, requires a graphical interface
│
├── test.py # Test accuracy after attack
│
├── test_num_workers.py # Test the optimal number of workers
│
├── train.py # Train recognition model
│
└── train_residual_model.py # Train the perturbation generation model for UPSET method
```
## Installation
```shell
git clone https://github.com/gralliry/Adversarial-Attack-Generation-Techniques.git
cd Adversarial-Attack-Generation-Techniques
pip install -r requirements.txt
```
## Train Recognition Model
```shell
# -e --epoch specifies the number of training epochs, optional, default is 100
python train.py -e 100
```
## Visualize Attack Effects
```shell
# -m specifies the attack method
# L-BFGS, FGSM, I-FGSM, JSMA, ONE-PIXEL, C&W, DEEPFOOL, MI-FGSM, UPSET
python contrast.py -m FGSM
```
## Test Accuracy After Attack
```shell
# -m specifies the attack method
# L-BFGS, FGSM, I-FGSM, JSMA, ONE-PIXEL, C&W, DEEPFOOL, MI-FGSM, UPSET
# -c specifies the number of test runs, optional, default is 500
python test.py -m FGSM -c 100
```
## Train Perturbation Generation Model for UPSET Method
```shell
# -t --target specifies the target label for the perturbation generation model
# Range is 0 ~ 9, corresponding to plane, automobile, bird, cat, deer, dog, frog, horse, ship, truck
# -e --epoch specifies the number of training epochs, optional, default is 100
# -lr --learning_rate specifies the learning rate, optional, default is 1e-3
# Training the perturbation model requires a recognition model, please modify or load it in the script
python train_residual_model.py -t 0 -e 100
```
## Test Optimal Number of Workers
```shell
# Run and select the best number of workers based on execution time
python test_num_workers.py
```
## Contributors
- Liang Jianye - SCAU