Utils to make adversarial experiments (classification task) easier, including natural training and testing, adversarial training and testing, visualizing.
Download this repository as a python package.
- Use tqdm bar as time monitor.
- Use dataclass to wrap hyper-parameters. (
dataclassesis new in Python 3.7) - Use type hint. (To be completed)
python >= 3.7
numpy == 1.19.2
torch == 1.7.0
torchvision == 0.8.1
matplotlib == 3.3.2
tqdm == 4.51.0
Adv-utils use torchvision.datasets to load data. The default data directory is ~/torchvision_dataset. You can change the variable config.data_info.dir_dataset for your own path.
The structure in the data directory should be like this:
├── cifar-10-batches-py
│ ├── batches.meta
│ ├── data_batch_1
│ ├── data_batch_2
│ ├── data_batch_3
│ ├── data_batch_4
│ ├── data_batch_5
│ ├── readme.html
│ └── test_batch
├── ILSVRC2012
│ ├── meta.bin
│ ├── train
│ └── val
├── MNIST
│ ├── processed
│ └── raw
├── ...- Natural:
demo/train.py - Adversarial:
demo/train_adv.py
-
Show some samples and do processing:
demo/vis_negative_samples.py
-
todo: Add more demos.
MadryLab - robustness