meet-cjli / ctrl Goto Github PK

python main_train.py --dataset cifar10 --mode normal --method simclr --threat_model our --channel 1 2 --trigger_position 15 31 --poison_ratio 0.01 --lr 0.06 --wd 0.0005 --magnitude 100.0 --poisoning --epochs 800 --gpu 0 --window_size 32 --trial clean

Use GPU: 0 for training => creating cnn model 'resnet18' CIFAR-variant Resnet is loaded Files already downloaded and verified Files already downloaded and verified

File "/workspace/ChengYi/CTRL/loaders/diffaugment.py", line 266, in set_aug_diff raise NotImplementedError NotImplementedError

How do I solve this?

Hi, thanks for presenting your great work here!
I successfully reproduce your result on cifar10 dataset with expected high ASR, around 80%.
However, when I am trying to transfer your attack on imagenet100 dataset (without down-sampling to 64*64), I use your code without any changes (window size=32, pos_list = [(15,15), (31,31)] magnitude=100) to generate poison samples in poisoning rate 1%, but the attack is unsuccessful with very low ASR. Do you have any suggestions? Thanks!

magnitude in the paper is 50 during training but got 100 in GitHub

I was wondering why number of classes is set to 100 for CIFAR10, in diffaugment.py function:set_aug_diff(args).

if args.dataset == 'cifar10':
    mean = (0.4914, 0.4822, 0.4465)
    std = (0.2023, 0.1994, 0.2010)
    args.size = 32
    args.num_classes = 100
    args.save_freq = 100

elif args.dataset == 'cifar100':
    mean = (0.5071, 0.4867, 0.4408)
    std = (0.2675, 0.2565, 0.2761)
    args.size = 32
    args.num_classes = 100
    args.save_freq = 100

elif args.dataset == 'imagenet-100':
    mean = (0.485, 0.456, 0.406)
    std = (0.229, 0.224, 0.225)
    args.size = 64
    args.save_freq = 100
    args.num_classes = 100

else:
    raise ValueError(args.dataset)

Dear authors, thank you for your wonderful work. I was wondering if it is possible to add a list of requirements to the repo, either in the form of requirements.txt or in the README.md.