princeton-sysml / gradattack Goto Github PK

Hi,

The requirements file lists Pytorch-lightning 1.5.1, however pip install Gradattack has pytorch-lightning 1.3.1 which has the deprecated manual backward being used. Can you please confirm which pytorch-lightning version to use?

Also, I've had issues replicating the baseline you achieved for the no defense method, where the reconstructions are very poor (even with a pretrained Resnet18 for 200 epochs, with 93% test accuracy, BN exact, and all the other recommended hyperparameters in your paper). Could I be missing an argument or something else?

In the documentation relative to the papers introducing gradient inversion attacks (GradAttack/papers/gradient_inversion.md) the GradInversion paper by Yin et al. lists "Good approximation of private labels" as additional information needed other than gradient.

The table entry is missing that the above paper also requires BatchNorm statistics (Section 3.3) as part of the regularization terms.

I noticed opacus is not supported batchnorm2d,so we should use convert_batchnorm_modules to convert batchnorm2d module to groupnorm. In this way, we can not use batchnorm statistics to conduct grad_attack, so how can solve this question. Thanks for your reply.

Having issues while testing the example script (side note the slack link is expired)

Versions:
Python 3.8.10
PyTorch 1.8.1+cu102
NumPy 1.21.4

python3 attack_cifar10_gradinversion.py --batch_size 16 --tv 0.05 --bn_reg 0.001 --reconstruct_labels

{'reconstruct_labels': True, 'signed_image': False, 'mini': False, 'large': False, 'BN_exact': False, 'attacker_eval_mode': False, 'defender_eval_mode': False, 'total_variation': 0.05, 'epoch': 0, 'bn_reg': 0.001, 'attack_lr': 0.1}
Global seed set to 1234
[ToTensor(), Normalize(mean=(0.4914, 0.4822, 0.4465), std=(0.2023, 0.1994, 0.201))]
Files already downloaded and verified
Files already downloaded and verified
Loaded data!
Traceback (most recent call last):
File "attack_cifar10_gradinversion.py", line 199, in
pipeline, attack_hparams = setup_attack()
File "attack_cifar10_gradinversion.py", line 77, in setup_attack
model = create_lightning_module(
File "/home/ubuntu/.local/lib/python3.8/site-packages/gradattack/models/init.py", line 493, in create_lightning_module
assert os.path.exists(ckpt), f"Failed to load checkpoint {ckpt}"
AssertionError: Failed to load checkpoint checkpoint/vanilla_epoch=1-step=1531.ckpt