type=int,
default=40,
help='number of attack iterations')
parser.add_argument('--resume', type=int, default=0)
parser.add_argument('--save_model_loc', type=str, default=None)
args = parser.parse_args()
print(args)
device = "cuda"
set_seed(0)
trainset, normalize, unnormalize = str2dataset(args.dataset, train=True)
trainloader = torch.utils.data.DataLoader(trainset,
batch_size=args.batch_size,
shuffle=True,
num_workers=2)
net = str2model(path=args.save_model_loc,
dataset=args.dataset,
pretrained=args.resume).eval().to(device)
if args.attack == "frank":
attacker = FrankWolfe(predict=lambda x: net(normalize(x)),
loss_fn=nn.CrossEntropyLoss(reduction="sum"),
eps=args.eps,
kernel_size=5,
nb_iter=args.nb_iter,
parser.add_argument('--save_info_loc', type=str_or_none, default=None)
parser.add_argument('--seed', type=int, default=0)
parser.add_argument('--postprocess', type=str2bool, default=False)
# parser.add_argument('--capacity_proj_mod', type=int, default=-1)
args = parser.parse_args()
print(args)
device = "cuda"
set_seed(args.seed)
testset, normalize, unnormalize = str2dataset(args.dataset)
testloader = torch.utils.data.DataLoader(testset,
batch_size=args.batch_size,
shuffle=False,
num_workers=0)
net = str2model(args.checkpoint, dataset=args.dataset,
pretrained=True).eval().to(device)
for param in net.parameters():
param.requires_grad = False
projected_gradient = ProjectedGradient(
predict=lambda x: net(normalize(x)),
loss_fn=nn.CrossEntropyLoss(reduction="sum"),
eps=args.eps,
