def main(args, io):
train_loader = DataLoader(S3DIS_cls(partition='train',
num_points=args.num_points,
test_area=args.test_area),
num_workers=8,
batch_size=args.batch_size,
shuffle=True,
drop_last=True)
test_loader = DataLoader(S3DIS_cls(partition='test',
num_points=args.num_points,
test_area=args.test_area),
num_workers=8,
batch_size=args.batch_size,
shuffle=False,
drop_last=False)
model = DGCNN(args)
model.cuda()
model = nn.DataParallel(model)
print("Let's use", torch.cuda.device_count(), "GPUs!")
opt = optim.SGD(model.parameters(),
lr=args.lr,
momentum=0.9,
weight_decay=1e-4)
scheduler = CosineAnnealingLR(opt, args.epochs)
best_test_loss = 1e10
for epoch in range(args.epochs):
io.cprint('Epoch [%d]' % (epoch + 1))
train(model, train_loader, opt, io)
scheduler.step()
test_loss = test(model, test_loader, io)
if test_loss < best_test_loss:
best_test_loss = test_loss
# save in torch==1.4 readible style
torch.save(model.state_dict(),
'checkpoints/%s/model_%s.t7' %
(args.exp_name, args.test_area),
_use_new_zipfile_serialization=False)
# load model weight
state_dict = torch.load(
BEST_WEIGHTS[args.model], map_location='cpu')
print('Loading weight {}'.format(BEST_WEIGHTS[args.model]))
try:
model.load_state_dict(state_dict)
except RuntimeError:
# eliminate 'module.' in keys
state_dict = {k[7:]: v for k, v in state_dict.items()}
model.load_state_dict(state_dict)
# distributed mode on multiple GPUs!
# much faster than nn.DataParallel
model = DistributedDataParallel(
model.cuda(), device_ids=[args.local_rank])
# setup attack settings
if args.adv_func == 'logits':
adv_func = LogitsAdvLoss(kappa=args.kappa)
else:
adv_func = CrossEntropyAdvLoss()
dist_func = L2Dist()
# hyper-parameters from their official tensorflow code
attacker = CWPerturb(model, adv_func, dist_func,
attack_lr=args.attack_lr,
init_weight=10., max_weight=80.,
binary_step=args.binary_step,
num_iter=args.num_iter)
# attack
print('Model not recognized')
exit(-1)
# load model weight
state_dict = torch.load(BEST_WEIGHTS[args.model], map_location='cpu')
print('Loading weight {}'.format(BEST_WEIGHTS[args.model]))
try:
model.load_state_dict(state_dict)
except RuntimeError:
# eliminate 'module.' in keys
state_dict = {k[7:]: v for k, v in state_dict.items()}
model.load_state_dict(state_dict)
# distributed mode on multiple GPUs!
# much faster than nn.DataParallel
model = DistributedDataParallel(model.cuda(), device_ids=[args.local_rank])
# setup attack settings
# budget, step_size, number of iteration
# settings adopted from CVPR'20 paper GvG-P
delta = args.budget
args.budget = args.budget * \
np.sqrt(args.num_points * 3) # \delta * \sqrt(N * d)
args.num_iter = int(args.num_iter)
args.step_size = args.budget / float(args.num_iter)
# which adv_func to use?
if args.adv_func == 'logits':
adv_func = LogitsAdvLoss(kappa=args.kappa)
else:
adv_func = CrossEntropyAdvLoss()
clip_func = ClipPointsL2(budget=args.budget)
print('Model not recognized')
exit(-1)
# load model weight
state_dict = torch.load(BEST_WEIGHTS[args.model], map_location='cpu')
print('Loading weight {}'.format(BEST_WEIGHTS[args.model]))
try:
model.load_state_dict(state_dict)
except RuntimeError:
# eliminate 'module.' in keys
state_dict = {k[7:]: v for k, v in state_dict.items()}
model.load_state_dict(state_dict)
# distributed mode on multiple GPUs!
# much faster than nn.DataParallel
model.cuda()
# model = torch.nn.DataParallel(
# model.cuda())
# setup attack settings
# budget, step_size, number of iteration
# settings adopted from CVPR'20 paper GvG-P
delta = args.budget
args.budget = args.budget * \
np.sqrt(args.num_points * 3) # \delta * \sqrt(N * d)
args.num_iter = int(args.num_iter)
args.step_size = args.budget / float(args.num_iter)
# which adv_func to use?
if args.adv_func == 'logits':
adv_func = LogitsAdvLoss(kappa=args.kappa)
else:
