dnn-security / watermark-robustness-toolbox Goto Github PK

I find a url (url = "https://www.dropbox.com/s/z11ds7jvewkgv18/adi.zip?dl=1") in trigger_datasets.py,but the url is invalid，anywhere else to find it? ANYONE CAN HELP ?THANK YOU!

Hi there,

Thanks for this great survey. It is informative and provides lots of insights. Thanks a lot for opensource this toolbox. However, when I tried to play with it, I encountered a ModuleNotFoundError.

I have installed all requirements and optional requirements. Then when I attempted to train a model on cifar10 using

python train.py --config configs/cifar10/train_configs/resnet.yaml

I encountered a ModuleNotFoundError as follows,

Traceback (most recent call last):
  File "train.py", line 14, in <module>
    from wrt.classifiers import PyTorchClassifier
  File "/PyWorkSpace/WRT/wrt/__init__.py", line 8, in <module>
    from wrt import attacks
  File "/PyWorkSpace/WRT/wrt/attacks/__init__.py", line 6, in <module>
    from wrt.attacks.removal.fine_tuning import FTLLAttack, FTALAttack, RTLLAttack, RTALAttack, FineTuningAttack
  File "/PyWorkSpace/WRT/wrt/attacks/removal/__init__.py", line 10, in <module>
    from .regularization import Regularization
  File "/PyWorkSpace/WRT/wrt/attacks/removal/regularization.py", line 37, in <module>
    from wrt.defenses import Watermark
  File "/PyWorkSpace/WRT/wrt/defenses/__init__.py", line 5, in <module>
    from backdoor import *
ModuleNotFoundError: No module named 'backdoor'

I've checked wrt/defenses/ and all possible places and could not find this backdoor.

I appreciate it if you can help me with this issue.

Best,
MENG2010

Thank you to the team for producing such a comprehensive toolkit.
I would like to ask if this kit can be used without a GPU, with a CPU verison torch.

hello，
thanks for your survey and the opensource for this toolbox, it helps me a lot for the understanding of watermarking schemes.

However, when i trained embed.py to embed a content watermark in to the cifar10 model, it throwed the following error :
"nll_loss_forward_reduce_cuda_kernel_2d_index" not implemented for 'float', it seems to occured when implementing the loss function nll_loss, but i have no idea how to solve it. Have you ever encountered this error when embedding watermarks? Any reply will be appreciated!