-
Notifications
You must be signed in to change notification settings - Fork 38
/
Copy pathvisualize.py
180 lines (151 loc) · 7.11 KB
/
visualize.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
import numpy as np
import matplotlib.pyplot as plt
import os
from utils import *
from index import *
from matplotlib import cm
from matplotlib.ticker import LinearLocator
import seaborn as sns
# sns.set_style("darkgrid")
sns.set_style("darkgrid", {"grid.color": ".6", "grid.linestyle": ":"})
Gini_index = Index()
def img_denorm(x, mean= [0.4914, 0.4822, 0.4465], std=[0.2023, 0.1994, 0.2010]):
# 3, H, W, B
ten = x.clone().permute(1, 2, 3, 0)
for t, m, s in zip(ten, mean, std):
t.mul_(s).add_(m)
# B, 3, H, W
return torch.clamp(ten, 0, 1).permute(3, 0, 1, 2)
def draw_perturb_loop(natural_perturb_tracer, img_id, result_dir):
plt.figure(figsize=(6,4))
tot_len = len(natural_perturb_tracer["confidence"][0])
if tot_len == (natural_perturb_tracer["correct_id"][0]):
plt.plot(np.arange(0,natural_perturb_tracer["correct_id"][0]), natural_perturb_tracer["confidence"][0], c = "b")
else:
correct_y = natural_perturb_tracer["confidence"][0][:natural_perturb_tracer["correct_id"][0]+1]
wrong_y = natural_perturb_tracer["confidence"][0][natural_perturb_tracer["correct_id"][0]:]
label_name = "Natural, " + "Std-" + str( "{:.4f}".format(natural_perturb_tracer["std"][0]))
plt.plot(np.arange(0,natural_perturb_tracer["correct_id"][0]+1), correct_y, c = "b")
plt.plot(np.arange(natural_perturb_tracer["correct_id"][0], tot_len), wrong_y, c = "r", linestyle = "-.")
plt.text(natural_perturb_tracer["correct_id"][0]-1, wrong_y[0]-1, 'X', horizontalalignment = "left", c="r", fontsize = 'x-large', fontweight = 'extra bold')
plt.ylabel("Confidence.")
plt.xlabel("Percentage of dropped pixels (‰).")
plt.ylim(0,100)
# plt.title("Confidence on target class.", fontweight='bold')
# plt.legend()
dir_path = os.path.join(result_dir, str(img_id))
if not os.path.exists(dir_path):
os.mkdir(dir_path)
plt.savefig(os.path.join(dir_path, "pixel_drop_check.png"))
plt.close()
def draw_normalized_importance_distribution(natural_img_gradshap, adv_img_gradshap, result_dir):
plt.figure(figsize=(12,3))
natural_gs_np = np.sort(natural_img_gradshap.cpu().detach().numpy().flatten())[::-1]
# natural_gs_np = natural_gs_np[natural_gs_np>0][::-1]
natural_gini = Gini_index.gini(abs(natural_gs_np))
natural_gs_np = natural_gs_np/np.max(natural_gs_np)
natural_gs_np = natural_gs_np[natural_gs_np>0]
# plt.fill_between( np.arange(len(aug_gs_np)), 0, aug_gs_np, color = "g", alpha=0.5, label='Aug.')
plt.fill_between( np.arange(len(natural_gs_np)), 0, natural_gs_np, color = "b", alpha = 0.7, label="Std. trained")
adv_gs_np = np.sort(adv_img_gradshap.cpu().detach().numpy().flatten())[::-1]
# natural_gs_np = natural_gs_np[natural_gs_np>0][::-1]
adv_gini = Gini_index.gini(abs(adv_gs_np))
adv_gs_np = adv_gs_np/np.max(adv_gs_np)
adv_gs_np = adv_gs_np[adv_gs_np>0]
plt.fill_between( np.arange(len(adv_gs_np)), 0, adv_gs_np, color = "r", alpha = 0.7, label="Adv. trained")
plt.xticks([])
plt.ylabel('Significance',fontweight='bold',fontsize = 12)
# plt.xlabel('Features (RGB)')
plt.title('Noramlized Features Attribution Distribution.', fontweight='bold', y = -0.2, fontsize = 14)
plt.legend(fontsize = 12, loc='center right')
# dir_path = os.path.join(result_dir, str(img_id))
# if not os.path.exists(dir_path):
# os.mkdir(dir_path)
plt.savefig(os.path.join(result_dir, "normalized_feature_gini.png"))
plt.close()
def draw_3d_importance_distribution(img, natural_img_gradshap, img_id, img_path):
feature_map_mean = np.max(natural_img_gradshap.cpu().detach().numpy()[0], axis = 0)
feature_map_positive = (feature_map_mean>0)*feature_map_mean
# Normalize
# feature_map_positive = feature_map_positive/np.max(feature_map_positive)
ax = plt.figure().add_subplot(projection='3d')
X = np.arange(feature_map_mean.shape[0])
Y = np.arange(feature_map_mean.shape[1])
X, Y = np.meshgrid(X, Y)
# Plot the surface.
surf = ax.plot_surface(X, Y, feature_map_mean, cmap=cm.coolwarm,
linewidth=0, antialiased=False)
# img_arr = np.moveaxis(img_denorm(img)[0].cpu().detach().numpy(), 0, -1)
# stride args allows to determine image quality
# stride = 1 work slow
# ax.plot_surface(X, Y, np.atleast_2d(-0.01), rstride=1, cstride=1, facecolors=img_arr)
ax.zaxis.set_major_locator(LinearLocator(10))
# A StrMethodFormatter is used automatically
ax.zaxis.set_major_formatter('{x:.02f}')
# alpha = 0.9
# surf.set_facecolor((0, 0, 1, alpha))
# Add a color bar which maps values to colors.
plt.colorbar(surf, shrink=0.5, aspect=5)
# ax.set_yticks([])
# ax.set_xticks([])
ax.set_zlim(-0.01, np.max(feature_map_positive))
ax.set_ylabel('y')
ax.set_xlabel('x')
ax.view_init(60, 35)
# dir_path = os.path.join(result_dir, str(img_id))
# if not os.path.exists(dir_path):
# os.mkdir(dir_path)
# plt.savefig(os.path.join(dir_path, "3d_feature_gini.png"))
plt.savefig(img_path)
plt.close()
def draw_org_img(input_img, img_path):
plt.figure(figsize = (6,6), frameon=False)
img = np.moveaxis(img_denorm(input_img)[0].cpu().detach().numpy(), 0, -1)
plt.imshow(img)
plt.yticks([])
plt.xticks([])
# dir_path = os.path.join(result_dir, str(img_id))
# if not os.path.exists(dir_path):
# os.mkdir(dir_path)
# plt.savefig(os.path.join(result_dir, "{}.png".format(str(img_id))))
plt.savefig(img_path, bbox_inches = 'tight', pad_inches= 0.05)
plt.close()
def draw_occ_img(input_img, title_name, img_path):
plt.figure(figsize = (6,6))
img = np.moveaxis(img_denorm(input_img)[0].cpu().detach().numpy(), 0, -1)
plt.imshow(img)
plt.yticks([])
plt.xticks([])
plt.title(title_name,y = -0.1, fontweight="bold", fontsize = 14)
# dir_path = os.path.join(result_dir, str(img_id))
# if not os.path.exists(dir_path):
# os.mkdir(dir_path)
# plt.savefig(os.path.join(result_dir, "{}.png".format(str(img_id))))
plt.savefig(img_path)
plt.close()
def draw_mask(input_mask, img_path):
plt.figure(figsize = (6,6))
img = np.moveaxis(input_mask[0].cpu().detach().numpy(), 0, -1)
plt.imshow(img)
plt.yticks([])
plt.xticks([])
# dir_path = os.path.join(result_dir, str(img_id))
# if not os.path.exists(dir_path):
# os.mkdir(dir_path)
# plt.savefig(os.path.join(result_dir, "{}.png".format(str(img_id))))
plt.savefig(img_path)
plt.close()
def draw_feature_map(feature_map, img_id, img_path):
plt.figure(figsize = (6,6))
feature_map = np.moveaxis(img_denorm(feature_map)[0].cpu().detach().numpy(), 0, -1)
feature_map = (feature_map > 0) * feature_map
feature_map = (feature_map/np.max(feature_map))*255.0
plt.imshow(feature_map)
plt.yticks([])
plt.xticks([])
# dir_path = os.path.join(result_dir, str(img_id))
# if not os.path.exists(dir_path):
# os.mkdir(dir_path)
# plt.savefig(os.path.join(result_dir, "{}.png".format(str(img_id))))
plt.savefig(img_path)
plt.close()