thuyimingli / backdoorbox Goto Github PK

你好，后门攻击的输出结果的含义有点不明白，输出如何查看攻击成功率呀

Traceback (most recent call last):
File "/home/objdet/Desktop/backdoor/BackdoorBox/example.py", line 66, in
badnets = core.BadNets(
TypeError: init() got an unexpected keyword argument 'poisoned_transform_index'

When I ran the code example.py, the above error was thrown out. And I have checked the code in core/attacks/BadNets, it does not have the argument poisoned_transform_index in function init().

Could you fix the bug?

Thanks.

If you encounter the following error: AssertionError: This machine has no cuda devices! To view your GPU number, run the "nvidia-smi" command and set CUDA_VISIBLE_DEVICES to the corresponding number at the beginning of the test file

你好，在BackdoorBox-main/BackdoorBox-main/tests/test_BadNets.py中，毒化数据是如何保存的呢？因为在SCALE-UP-main/SCALE-UP-main/test_BadNets.py文件中保存的毒化样本似乎有着格式问题。

你好！如题。

Hi, this is some amazing work and is really helpful for people just starting out. I think I might have found a bug and wanted to let you know.

In attacks -> WaNET.py -> add_trigger(self, img, noise=False)

• If there is noise, the grid is calculated as: grid = self.grid + ins / self.h
• However, in the next line grid is recalculated as: grid = torch.clamp(self.grid + ins / self.h, -1, 1)

Im also attaching an image

Hi big brother, I'm a graduate student from sjtu doing research on backdoor learning, thank you for the project it is of great help to my study. I've experimented with different timing of injecting backdoor trigger in torchvision transforms previously, and I'm curious that do you have empirical results on the effect of injecting backdoor triggers at different stage of image augmentation?

Thank you very much for your time.

The script "test_SleeperAgent.py" with cifar10 dataset achieves only 9.99% ASR after poisoning:

Epoch 100[2023-06-01_00:45:58] train_acc: 99.80, test_acc: 93.11, source_asr: 8.60, full_asr: 9.99

当我想使用多 GPU 时，我将 scheduler 做了如下改动：'CUDA_VISIBLE_DEVICES': '0, 1', 'GPU_num': 2，然后出现了如下错误：

Traceback (most recent call last):
  File "example.py", line 146, in <module>
    badnets.train(schedule)
  File "/BackdoorBox-main/core/attacks/base.py", line 167, in train
    self.model = self.model.to(device)
UnboundLocalError: local variable 'device' referenced before assignment

我认为这个错误出现在使用多个 GPU 时，没有将 device 指定为 GPU，导致程序后续在移动数据到指定设备上时，发现之前并没有指定 device。
我的解决方案是添加指定 device 的一行代码，截图如下：

我不知道这样的解决方案是否是最佳的。我期待你们修复这个 bug 🥂

Hi, thanks for the great work!
What is the adversarial model you used for label consistent attack?