if __name__ == "__main__":
parser = argparse.ArgumentParser(description='ori_model')
parser.add_argument("model", type=str, help="ori_model")
parser.add_argument("-eps", type=int, help="eps")
parser.add_argument("-alpha", type=int, help="alpha")
parser.add_argument("-iters", type=int, help="iters")
parser.add_argument("-out", type=int, help="out")
parser.add_argument("-epochs", type=int, help="epochs")
args = parser.parse_args()
torch.manual_seed(123456)
dataloaders, dataset_sizes = data_process(batch_size=64)
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model_ft = VGG_16()
model_ft.load_weights() #use this to do the adversarial training
#model_ft.load_state_dict(torch.load('../donemodel/'+args.model))
#use this to do the curriculum adversarial training
model_ft.to(device)
# model_ft = nn.DataParallel(model,device_ids=[0,1])
print("eps is ", args.eps, " ", "alpha is ", args.alpha, " ",
"iteration is ", args.iters, " ", "out is", args.out, " ",
"epochs is ", args.epochs)
criterion = nn.CrossEntropyLoss()
optimizer_ft = optim.Adam(model_ft.parameters(), lr=0.0001)
parser.add_argument("--N",
type=int,
default=1000,
help="number of samples to use")
parser.add_argument("--alpha",
type=float,
default=0.001,
help="failure probability")
args = parser.parse_args()
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
f = open(args.outfile, 'w')
print("idx\tlabel\tpredict\tcorrect\ttime", file=f, flush=True)
batch_size = 1
dataloaders, dataset_sizes = data_process(batch_size)
model = VGG_16()
model.load_state_dict(torch.load('../donemodel/' + args.model))
model.to(device)
smoothed_classifier = Smooth(model, 10, args.sigma)
eps = [0.5, 1, 1.5, 2, 2.5, 3] # eps is epsilon of the l_2 bound
alpha = [0.05, 0.1, 0.15, 0.2, 0.25, 0.3] # alpha is learning rate
itera = [20, 20, 20, 20, 20, 20] # iterations to find optimal
restart = [
1, 1, 1, 1, 1, 1
] # restart times, since we just do some standard check of our model,
# we do not use mutliple restarts, but you can change that if you want
# delete some hyperparmeters could speed up
for i in range(len(eps)):
cor = 0
#patch_size = opt.patch_size
image_size = opt.image_size
plot_all = opt.plot_all
print("=> creating model ")
netClassifier_par = {
'input_size': [3, 224, 224],
'input_range': [0, 255],
'mean':
[0.367035294117647, 0.41083294117647057, 0.5066129411764705],
'std': [1 / 255, 1 / 255, 1 / 255],
'num_classes': 10,
'input_space': "RGB"
}
netClassifier = VGG_16()
netClassifier.load_state_dict(torch.load('../donemodel/' + opt.model))
if torch.cuda.is_available():
netClassifier.cuda()
print('==> Preparing data..')
data_dir = '../Data'
normalize = transforms.Normalize(
mean=[0.367035294117647, 0.41083294117647057, 0.5066129411764705],
std=[1 / 255, 1 / 255, 1 / 255])
train_loader = torch.utils.data.DataLoader(dset.ImageFolder(
os.path.join(data_dir, 'Train_patch'),
transforms.Compose([
transforms.Resize(