Adversarial-Defenses-PyTorch (Under Reconstruction)
[Torchdefenses] is a PyTorch library that provides adversarial defenses to obtain robust model against adversarial attacks. It contains PyTorch Lightning-like interface and functions that make it easier for PyTorch users to implement adversarial defenses.
How to use?
import torchdefenses as td
rmodel = td.RobModel(model, n_classes=10,
normalize={'mean':[0.4914, 0.4822, 0.4465], 'std':[0.2023, 0.1994, 0.2010]})
Easy training
import torchdefenses.trainer as tr
trainer = tr.Standard(rmodel)
trainer.record_rob(train_loader, val_loader, eps=0.3, alpha=0.1, steps=5, std=0.1)
trainer.fit(train_loader=train_loader, max_epoch=10, optimizer="SGD(lr=0.01)",
scheduler="Step([100, 105], 0.1)", scheduler_type="Epoch",
record_type="Epoch", save_type="Epoch",
save_path="./_temp/"+"sample", save_overwrite=True)
Supporting Multi-GPU
import os
os.environ["CUDA_VISIBLE_DEVICES"] = "1,2" # Possible GPUS
model = td.utils.load_model(model_name="ResNet18", n_classes=10).cuda() # Load model
model = torch.nn.DataParallel(model) # Parallelize
rmodel = td.RobModel(model, n_classes=10,
normalize={'mean':[0.4914, 0.4822, 0.4465], 'std':[0.2023, 0.1994, 0.2010]}) # Wrap it with RobModel
trainer = ... # Define trainer
trainer.fit(..) # Training start
Recording, Saving and Visualizing
trainer.save_all("./_temp/"+"sample", overwrite=True)
trainer.rm.plot(title="A", xlabel="Epoch", ylabel="Accuracy",
figsize=(6, 4),
x_key='Epoch',
y_keys=['Clean(Tr)', 'FGSM(Tr)', 'PGD(Tr)', 'GN(Tr)',
'Clean(Val)', 'FGSM(Val)', 'PGD(Val)', 'GN(Val)'],
ylim=(-10, 110),
colors=['k', '#D81B60', '#1E88E5', '#004D40']*2,
labels=['Clean', 'FGSM', 'PGD', 'GN', '', '', '', ''],
linestyles=['-', '-', '-', '-', '--', '--', '--', '--'],
)
Easy evaluation
rmodel.eval_accuracy(test_loader)
rmodel.eval_rob_accuracy_pgd(test_loader, eps=1, alpha=0.1,
steps=10, random_start=True, restart_num=1)
Useful functions
from torchdefenses.utils import fix_randomness, fix_gpu
fix_randomness(0)
fix_gpu(0)
How to customize?
1 maintainer
Metadata
1 maintainer
Release files for torchdefenses 0.1.25
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| torchdefenses-0.1.25-py3-none-any.whl | Python 3 | none | any | Details for torchdefenses-0.1.25-py3-none-any.whl |
Release files / torchdefenses-0.1.25-py3-none-any.whl
| Download URL | torchdefenses-0.1.25-py3-none-any.whl |
|---|---|
| Size | 41.7 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
17a82d37d4aa404711ffae172c39fd703e24f8d62c4bce64219d3b8ab7a4947c
|
|
BLAKE2b-256 checksum How to use checksums |
92a3f9cc6c3aac2f45333ca6d2fd48f3309985256f929bb194a24708a51042de
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/3.4.1 importlib_metadata/3.10.0 pkginfo/1.7.0 requests/2.25.1 requests-toolbelt/0.9.1 tqdm/4.59.0 CPython/3.8.8
|
Security for torchdefenses 0.1.25
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.
