-
Notifications
You must be signed in to change notification settings - Fork 0
/
test_utils.py
189 lines (168 loc) · 7.48 KB
/
test_utils.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
import matplotlib.pyplot as plt
import numpy as np
import skimage.measure
import os
from scipy.ndimage.filters import gaussian_filter
import seaborn as sns
import heapq
def plt_densitymap(img,adv,index,name):
npimg = img.cpu().numpy()
npadv = adv.cpu().numpy()
fig = plt.figure()
ax1 = fig.add_subplot(111)
ax1.axes.get_xaxis().set_visible(False)
ax1.axes.get_yaxis().set_visible(False)
perturb = np.abs(npadv-npimg)
perturb = np_sparsity(perturb)
perturb_smooth = gaussian_filter(perturb,sigma=1)
sns.heatmap(perturb_smooth, vmin=np.min(perturb_smooth), vmax=np.max(perturb_smooth), cmap ="coolwarm" , cbar=True )
if not os.path.isdir('example_visualize/img{}/'.format(index)):
os.mkdir('example_visualize/img{}/'.format(index))
plt.savefig('example_visualize/img{}/{}.png'.format(index,name))
return perturb_smooth
def plt_heatmap(img,adv,index,name):
npimg = img.cpu().numpy()
npadv = adv.cpu().numpy()
fig = plt.figure()
ax = fig.add_subplot(111)
ax.axes.get_xaxis().set_visible(False)
ax.axes.get_yaxis().set_visible(False)
perturb = np.abs(npadv-npimg)
perturb = np_sparsity(perturb)
np_min = np.min(perturb)
np_max = np.max(perturb)
np_int = np_max-np_min
im.set_clim(np_min+0.1*np_int,np_max-0.1*np_int)
cbar = fig.colorbar(ax=ax, mappable=im, orientation='vertical')
if not os.path.isdir('example_visualize/img{}/'.format(index)):
os.mkdir('example_visualize/img{}/'.format(index))
plt.savefig('example_visualize/img{}/{}.png'.format(index,name))
def plt_heatmap_compare(img,denoised,adv,index,name):
npimg = img.cpu().numpy()
npdenoised = denoised.cpu().numpy()
npadv = adv.cpu().numpy()
fig = plt.figure()
ax = fig.add_subplot(121)
bx = fig.add_subplot(122)
ax.axes.get_xaxis().set_visible(False)
ax.axes.get_yaxis().set_visible(False)
bx.axes.get_xaxis().set_visible(False)
bx.axes.get_yaxis().set_visible(False)
perturb = np.abs(npadv-npimg)
perturb = np_sparsity(perturb)
perturb_recon = np.abs(npadv-npdenoised)
perturb_recon = np_sparsity(perturb_recon)
im = ax.imshow(perturb,cmap='YlGnBu')
im_recon = bx.imshow(perturb_recon,cmap='YlGnBu')
np_min = np.min(perturb)
np_max = np.max(perturb)
np_int = np_max-np_min
np_min_recon = np.min(perturb_recon)
np_max_recon = np.max(perturb_recon)
np_int_recon = np_max_recon-np_min_recon
im.set_clim(np_min+0.1*np_int,np_max-0.1*np_int)
im_recon.set_clim(np_min_recon+0.1*np_int_recon,np_max-0.1*np_int_recon)
cbar = fig.colorbar(ax=ax, mappable=im, orientation='vertical')
cbar = fig.colorbar(ax=bx, mappable=im_recon, orientation='vertical')
if not os.path.isdir('example_visualize/img{}/'.format(index)):
os.mkdir('example_visualize/img{}/'.format(index))
plt.savefig('example_visualize/img{}/{}{}.png'.format(index,name,index))
def np_sparsity(array):
# reduce image dimension from 3 to 1
array = (array**2).sum(axis=0)
array = skimage.measure.block_reduce(array, (2,2), np.sum)
array = np.sqrt(array)
# print(array.shape)
return array
def hist_compare(img,denoised,adv,index,name):
npimg = img.cpu().numpy().ravel()
npdenoised = denoised.cpu().numpy().ravel()
npadv = adv.cpu().numpy().ravel()
pixel_ind = np.arange(len(npimg))
perturb_gt = npadv-npimg # a 1-dim flattened perturbation
perturb_rec = npadv-npdenoised # a 1-dim flattened reconstruced perturbation
fig = plt.figure()
plt.plot(pixel_ind,perturb_gt,label = 'ground truth perturbation')
plt.plot(pixel_ind,perturb_rec,label = 'reconstructed perturbation')
plt.xlabel('pixels')
plt.ylabel('perturbation')
plt.legend()
plt.title('groundtruth perturbation vs reconstructed perturbation')
plt.savefig('example_visualize/img{}/{}_pixelwise.png'.format(index,name))
plt.close(fig)
def hist_compare_few(img,denoised,adv,index,name):
npimg = img.cpu().numpy().ravel()
npdenoised = denoised.cpu().numpy().ravel()
npadv = adv.cpu().numpy().ravel()
pixel_ind = np.arange(len(npimg))
perturb_gt = npadv-npimg # a 1-dim flattened perturbation
perturb_rec = npadv-npdenoised # a 1-dim flattened reconstruced perturbation
fig = plt.figure()
plt.plot(pixel_ind[0:100],perturb_gt[0:100],label = 'ground truth perturbation')
plt.plot(pixel_ind[0:100],perturb_rec[0:100],label = 'reconstructed perturbation')
plt.xlabel('pixels')
plt.ylabel('perturbation')
plt.legend()
plt.title('groundtruth perturbation vs reconstructed perturbation')
plt.savefig('example_visualize/img{}/{}_pixelwise.png'.format(index,name))
plt.close(fig)
def distances_l1l2(img1,img2):
npimg1 = img1.cpu().numpy().ravel()
npimg2 = img2.cpu().numpy().ravel()
l1_distance = np.sum(np.abs(npimg1-npimg2))
l2_distance = np.sqrt(np.square(npimg1-npimg2))
return l1_distance, l2_distance
def IoU(perturb1,perturb2,top_percent):
perturb1 = (perturb1.cpu().numpy()**2).sum(axis=0).flatten()
perturb2 = (perturb2.cpu().numpy()**2).sum(axis=0).flatten()
# print(np.min(heapq.nlargest(int(len(perturb1)*top_percent),perturb1)))
mask_perturb1 = perturb1 >= np.min(heapq.nlargest(int(len(perturb1)*top_percent),perturb1))
mask_perturb2 = perturb2 >= np.min(heapq.nlargest(int(len(perturb2)*top_percent),perturb2))
I = np.sum(mask_perturb1 * mask_perturb2)
U = np.sum(mask_perturb1) + np.sum(mask_perturb2) - I
return I/U
def IoUplot(perturb1,perturb2, img, top_percent):
perturb1 = (perturb1.cpu().numpy()**2).sum(axis=0).flatten()
mask_perturb1 = perturb1 >= np.min(heapq.nlargest(int(len(perturb1)*top_percent),perturb1))
fig = plt.figure()
plt.imshow(np.transpose(img.cpu().numpy(), (1, 2, 0)))
plt.imshow(mask_perturb1.reshape(32,32),alpha=0.3)
plt.savefig('example_visualize/gt_mask.png')
perturb2 = (perturb2.cpu().numpy()**2).sum(axis=0).flatten()
mask_perturb2 = perturb2 >= np.min(heapq.nlargest(int(len(perturb2)*top_percent),perturb2))
fig = plt.figure()
plt.imshow(np.transpose(img.cpu().numpy(), (1, 2, 0)))
plt.imshow(mask_perturb2.reshape(32,32),alpha=0.3)
plt.savefig('example_visualize/recon_mask.png')
def PSR(clean, adv, classifier, name):
# print(torch.unsqueeze(clean,0).shape)
t0 = classifier(torch.unsqueeze(clean,0)).argmax(1).detach().clone()
t = classifier(torch.unsqueeze(adv,0)).argmax(1).detach().clone()
_, length, width = clean.shape
d0 = np.zeros([length,width])
dt = np.zeros([length,width])
Z_adv = classifier(torch.unsqueeze(adv,0))
Z_adv_t0 = Z_adv[0][t0]
# print(Z_adv.shape)
Z_adv_t = Z_adv[0][t]
for i in range(length):
for j in range(width):
ablation = adv
ablation[:,i,j] = clean[:,i,j]
Z_ablation = classifier(torch.unsqueeze(ablation,0))
Z_ablation_t0 = Z_ablation[0][t0].detach().clone()
Z_ablation_t = Z_ablation[0][t].detach().clone()
# print(Z_ablation_t0-Z_adv_t0)
d0[i][j] = np.max([Z_ablation_t0-Z_adv_t0,0.01])
dt[i][j] = np.max([Z_adv_t-Z_ablation_t,0.01])
PSR = np.log2(dt/d0)
# S = dt+d0
# S = S >= np.min(heapq.nlargest(int(len(S.flatten())*0.7),S.flatten()))
PSR = PSR>=1
# print(PSR)
fig,ax=plt.subplots()
npimg = clean.cpu().numpy() # convert from tensor
ax.imshow(np.transpose(npimg, (1, 2, 0)))
ax.imshow(PSR,alpha=0.5,cmap="gray")
plt.savefig('example_visualize/PSR_{}.png'.format(name))
plt.close(fig)