-
Notifications
You must be signed in to change notification settings - Fork 1
/
evaluate_fashionAdv_attack.py
293 lines (219 loc) · 11.6 KB
/
evaluate_fashionAdv_attack.py
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
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
import os
import torch
import numpy as np
import random
from tqdm import tqdm
import time
import itertools
import shutil
import pickle
from pycocotools.coco import COCO
from pycocotools.cocoeval import COCOeval
from torchvision.utils import save_image
from torchvision import transforms
from torch.nn.functional import interpolate
import sys
from pytorch_msssim import ssim
from fashionadv_utils.util import normalize_imagenet, get_item
sys.path.extend(['./', './yolact_package'])
from yolact_package.evaluate_colorfool import *
from yolact_package.eval import *
from yolact_package.data import *
# ---------------- Configuration ----------------
with open('./configuration.json', 'r') as f:
configuration = json.load(f)
device = torch.device("cuda")
cudnn.fastest = True
torch.set_default_tensor_type('torch.cuda.FloatTensor')
seed = 0
random.seed(seed)
os.environ['PYTHONHASHSEED'] = str(seed)
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
torch.backends.cudnn.enabled = False
torch.backends.cudnn.benchmark = False
torch.backends.cudnn.deterministic = True
cocoGt = COCO(f"{configuration['annotations']}/instances_val2017.json")
set_cfg("yolact_base_config")
dataset = COCODetection(f"{configuration['val2017']}/", f"{configuration['annotations']}/instances_val2017.json",
transform=NoneTransform(), has_gt=True)
conv_id_coco_dset = {dataset.ids[i]: i for i in range(len(dataset))}
prep_coco_cats()
yolact = Yolact()
yolact.to(device)
yolact.load_weights(configuration['weights_yolact'])
yolact.eval()
for param in yolact.parameters():
param.requires_grad = False
args = parse_args('')
args.top_k = 5
cfg.mask_proto_debug = args.mask_proto_debug
yolact.detect.use_fast_nms = True
yolact.detect.use_cross_class_nms = False
cfg.train_masks = True
cfg.use_semantic_segmentation_loss = True
cfg.mask_proto_loss = 'l1'
cfg.pred_outs = False
conv_id_coco_dset = {dataset.ids[i]: i for i in range(len(dataset))}
# The id of images are formate with the id of the coco dataset
file_id = open(configuration['image_id_list'], 'rb')
imgs_to_attack_cocoid = pickle.load(file_id)
file_id.close()
imgs_to_attack = [conv_id_coco_dset[i] for i in imgs_to_attack_cocoid]
def generate_coco_file(dataset_indices, name, results, dataset, net, qf=-1):
"""This function generated the bbox and the mask files that follow the rules
of COCOapi.
Before using this function make sure that args.bbox_det_file and
args.mask_det_file are well define.
"""
if name not in results:
results[name] = {}
detections = Detections()
list_naturalness = []
for image_idx in tqdm(dataset_indices):
_, _, _, h, w, _ = dataset.pull_item(image_idx)
img_gt_open = Image.open(f'{configuration["result"]}/clean/{dataset.ids[image_idx]:012d}.png')
img_gt = transforms.ToTensor()(img_gt_open).to(device)
compression = '' if qf == -1 else f'_compress_{qf}'
img_adv = Image.open(f'{configuration["result"]}/adversarials/adversarial_{dataset.ids[image_idx]:012d}{compression}.png')
img_adv = transforms.ToTensor()(img_adv).to(device)
denom = torch.sqrt(torch.mean(img_adv * img_adv))
mse = torch.mean((img_adv - img_gt) ** 2)
nrmse = torch.sqrt(mse) / denom
psnr = - 10 * torch.log10(mse + 1e-10)
ssim_score = ssim(img_adv.unsqueeze(0), img_gt.unsqueeze(0), data_range=1.)
if qf is not -1:
img_gt_open.save(f'{configuration["result"]}/clean/temp_comp.png', format='JPEG', quality=qf)
img_gt_c = Image.open(f'{configuration["result"]}/clean/temp_comp.png')
img_gt_c = transforms.ToTensor()(img_gt_c).to(device)
mse_c = torch.mean((img_adv - img_gt_c) ** 2)
nrmse_c = torch.sqrt(mse_c) / denom
psnr_c = - 10 * torch.log10(mse_c + 1e-10)
ssim_score_c = ssim(img_adv.unsqueeze(0), img_gt_c.unsqueeze(0), data_range=1.)
list_naturalness.append((ssim_score.item(), mse.item(), nrmse.item(), psnr.item(),
ssim_score_c.item(), mse_c.item(), nrmse_c.item(), psnr_c.item()))
else:
list_naturalness.append((ssim_score.item(), mse.item(), nrmse.item(), psnr.item()))
img_adv = normalize_imagenet(img_adv)
preds = net(img_adv.unsqueeze(0))
my_prep_metrics(preds, h, w, dataset.ids[image_idx], detections)
detections.dump()
lenght = len(list_naturalness)
results[name].update({"ssim": sum([i[0] for i in list_naturalness]) / lenght,
"mse": sum([i[1] for i in list_naturalness]) / lenght,
"nrmse": sum([i[2] for i in list_naturalness]) / lenght,
"psnr": sum([i[3] for i in list_naturalness]) / lenght})
if qf is not -1:
results[name].update({"ssim_comp": sum([i[4] for i in list_naturalness]) / lenght,
"mse_comp": sum([i[5] for i in list_naturalness]) / lenght,
"nrmse_comp": sum([i[6] for i in list_naturalness]) / lenght,
"psnr_comp": sum([i[7] for i in list_naturalness]) / lenght})
def evaluate_COCOapi(list_id_image, name, results, option, cocoGt):
"""
With the coco API calculate the AP.
"""
if name not in results:
results[name] = {}
for t in ['mask', 'bbox']:
cocomask = cocoGt.loadRes(option[f'{t}_file_name'])
cocoEval = COCOeval(cocoGt, cocomask, '{}'.format('segm' if t is 'mask' else t))
cocoEval.params.imgIds = [dataset.ids[i] for i in list_id_image]
cocoEval.params.catIds = [1]
cocoEval.evaluate()
cocoEval.accumulate()
cocoEval.summarize()
results[name].update({f'AP_{t}': cocoEval.stats[0], f'AP50_{t}': cocoEval.stats[1],
f'AP75_{t}': cocoEval.stats[2], f'AP_small_{t}': cocoEval.stats[3],
f'AP_medium_{t}': cocoEval.stats[4], f'AP_large_{t}': cocoEval.stats[5],
f'AR1_{t}': cocoEval.stats[6], f'AR10_{t}': cocoEval.stats[7],
f'AR100_{t}': cocoEval.stats[8], f'AR_small_{t}': cocoEval.stats[9],
f'AR_medium_{t}': cocoEval.stats[10], f'AR_large_{t}': cocoEval.stats[11]})
def print_result(results, end=', '):
print("JPEG Quality Factor")
for t in ['mask', 'bbox']:
for m in ['AP_', 'AP50_']:
print('{} : {}'.format(t, 'mAP ' if m == 'AP_' else 'mAP50'), end=" = ")
for i in ['All_compress_1', 'All_compress_10', 'All_compress_20', 'All_compress_30', 'All_compress_40', 'All_compress_50', 'All_compress_60', 'All_compress_70', 'All_compress_80', 'All_compress_90', 'All_compress_100', 'All']:
print(f"{results[i][f'{m}{t}']*100:.5f}", end=end)
print()
print()
for t in ['ssim', 'ssim_comp', 'psnr', 'psnr_comp', 'mse', 'mse_comp', 'nrmse', 'nrmse_comp']:
print(f'{t}')
for i in ['All_compress_1', 'All_compress_10', 'All_compress_20', 'All_compress_30', 'All_compress_40',
'All_compress_50', 'All_compress_60', 'All_compress_70', 'All_compress_80', 'All_compress_90',
'All_compress_100']:
print(f"{results[i][t]:.7f}", end=end)
if t[-5:] == '_comp':
print(f"{results['All'][t[:-5]]:.7f}", end=end)
else:
print(f"{results['All'][t]:.7f}", end=end)
print()
print()
def generate_adversarial_image(dataset_list):
for content_index_image in tqdm(dataset_list):
clean_image_save = get_item(content_index_image, dataset, normalize=False)[0]
save_image(clean_image_save, f'{configuration["result"]}/clean/{dataset.ids[content_index_image]:012d}.png')
clean_image = get_item(content_index_image, dataset, normalize=False)[0].to(device)
image_mask = Image.open(f'{configuration["mask_upper_shirt"]}/mask_{dataset.ids[content_index_image]:012d}.png')
mask = transforms.ToTensor()(image_mask).to(device)
image_patch = Image.open(f'{configuration["result"]}/patch_{dataset.ids[content_index_image]:012d}.png')
patch = transforms.ToTensor()(image_patch).to(device)
adversarial_img = torch.mul((1 - mask), clean_image) + torch.mul(mask, patch)
save_image(adversarial_img, f'{configuration["result"]}/adversarials/adversarial_{dataset.ids[content_index_image]:012d}.png')
def generate_compress_adversarial_image(dataset_list, quality_factor):
"""
Generate the compress images
"""
for content_index_image in dataset_list:
adversarial_img = Image.open(f'{configuration["result"]}/adversarials/adversarial_{dataset.ids[content_index_image]:012d}.png')
compress_path = f'{configuration["result"]}/adversarials/adversarial_{dataset.ids[content_index_image]:012d}_compress_{quality_factor}.png'
adversarial_img.save(compress_path, format='JPEG', quality=quality_factor)
def supress_compress_adversarial_image(dataset_list, quality_factor):
for content_index_image in dataset_list:
compress_path = f'{configuration["result"]}/adversarials/adversarial_{dataset.ids[content_index_image]:012d}_compress_{quality_factor}.png'
os.remove(compress_path)
def my_evaluation(dataset_indices, option, dset=dataset, net=yolact, cocoGt=cocoGt):
"""
This function perfome the Evaluation with
"""
results = dict()
generate_coco_file(dataset_indices, 'All', results, dset, net)
# COCO Evaluation on uncompress images
evaluate_COCOapi(dataset_indices, 'All', results, option, cocoGt)
# JPEG Evaluation
for quality_factor in tqdm([1, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100]):
generate_compress_adversarial_image(dataset_indices, quality_factor)
option['bbox_file_name'] = f'{configuration["result"]}/evaluation/bbox_detections_compress_{quality_factor}.json'
option['mask_file_name'] = f'{configuration["result"]}/evaluation/mask_detections_compress_{quality_factor}.json'
args.bbox_det_file = option['bbox_file_name']
args.mask_det_file = option['mask_file_name']
# Compress images
generate_coco_file(dataset_indices, f'All_compress_{quality_factor}', results, dset, net, quality_factor)
# Evaluate compress images
evaluate_COCOapi(dataset_indices, f'All_compress_{quality_factor}', results, option, cocoGt)
# Suppress compress images
supress_compress_adversarial_image(dataset_indices, quality_factor)
return results
if __name__ == '__main__':
try:
os.mkdir(f'{configuration["result"]}/evaluation')
os.mkdir(f'{configuration["result"]}/adversarials')
os.mkdir(f'{configuration["result"]}/clean')
generate_adversarial_image(imgs_to_attack)
args.bbox_det_file = f'{configuration["result"]}/evaluation/bbox_detections.json'
args.mask_det_file = f'{configuration["result"]}/evaluation/mask_detections.json'
op = {'bbox_file_name': args.bbox_det_file, 'mask_file_name': args.mask_det_file}
results = my_evaluation(imgs_to_attack, op)
original_stdout = sys.stdout # Save a reference to the original standard output
with open(f'{configuration["result"]}/evaluation/results.txt', 'w') as f:
sys.stdout = f # Change the standard output to the file we created.
print_result(results, '\n')
sys.stdout = original_stdout
file_res = open(f'{configuration["result"]}/evaluation/results.pkl', 'wb')
pickle.dump(results, file_res)
file_res.close()
shutil.rmtree(f'{configuration["result"]}/clean', ignore_errors=True)
except FileExistsError:
print('already done')