Repository navigation
Expand file tree
/
Copy pathrun_detector.py
More file actions
87 lines (68 loc) · 3.14 KB
/
Copy pathrun_detector.py
File metadata and controls
87 lines (68 loc) · 3.14 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
import os
import torch
import copy
import argparse
from utils import load_model, load_dataloader, SPLIT_SYM, WORK_DIR, DETECTOR_RES_DIR
from src.dlcl import TargetDevice
from utils import DETECTOR_LIST
def _load_our_model(save_dir):
possible_list = [str(5 * i + 4) + '.tar' for i in range(20)]
possible_list = list(reversed(possible_list))
possible_list = ["best.tar"] + possible_list
save_path_list = [os.path.join(save_dir, path_name) for path_name in possible_list]
is_load = False
bd_trigger, save_model = None, None
for save_path in save_path_list:
if os.path.exists(save_path):
[bd_trigger, save_model, _] = torch.load(
save_path, map_location=torch.device('cpu'), weights_only=False
)
is_load = save_path
break
return is_load, bd_trigger, save_model
def get_our_score(detector, model_data_name, clean_model, hardware_id, cl_id):
hardware_target = TargetDevice(hardware_id)
task_name = model_data_name + SPLIT_SYM + f"CL___{cl_id}" + SPLIT_SYM + str(hardware_target)
save_dir = os.path.join(WORK_DIR, task_name)
is_load, our_bd_trigger, our_model = _load_our_model(save_dir)
if is_load:
our_model.init()
our_model.class_num = clean_model.class_num
our_score = detector.compute_score(our_model, our_bd_trigger)
return our_score, our_bd_trigger
def main(task_id, detector_id):
batch_size = 200
device = torch.device('cuda')
detector_cls = DETECTOR_LIST[detector_id]
train_loader, valid_loader, test_loader = load_dataloader(task_id, False, batch_size, batch_size)
detector = detector_cls(valid_loader, test_loader, device)
clean_model = load_model(task_id, load_pretrained=True)
model_data_name = clean_model.model_data_name
belt_save_dir = os.path.join("model_weight", "belt")
belt_model = copy.deepcopy(clean_model)
[belt_bd_trigger, state_dict] = torch.load(os.path.join(belt_save_dir, f"{model_data_name}."))
belt_model.load_state_dict(state_dict)
# hardware_id, cl_id
belt_score = detector.compute_score(belt_model, belt_bd_trigger)
final_res = {
"belt_score": belt_score
}
cl_score = detector.compute_score(clean_model, belt_bd_trigger)
final_res["cl_score"] = cl_score
for hardware_id in [-1, 0]:
for cl_id in range(3):
try:
our_score, our_bd_trigger = get_our_score(detector, model_data_name, clean_model, hardware_id, cl_id)
except Exception as e:
our_score, our_bd_trigger = None, belt_bd_trigger
final_res[f'ours_hardware_{hardware_id}_cl_{cl_id}'] = our_score
task_dir = os.path.join(DETECTOR_RES_DIR, str(task_id))
os.makedirs(task_dir, exist_ok=True)
torch.save(final_res, os.path.join(task_dir, f"{detector.__class__.__name__}.det"))
print(task_id, detector_id, 'successful')
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--task_id', type=int, default=0)
parser.add_argument('--detector_id', type=int, default=1)
args = parser.parse_args()
main(task_id=args.task_id, detector_id=args.detector_id)