-
Notifications
You must be signed in to change notification settings - Fork 3
/
plot_results_cifar10.py
149 lines (128 loc) · 5.45 KB
/
plot_results_cifar10.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
import os
import matplotlib.pyplot as plt
import torch
from utils import robust_accuracy_curve
result_dir = os.path.join('results, cifar10')
os.makedirs(os.path.join('results', 'curves'), exist_ok=True)
plt.rc('text', usetex=True)
plt.rc('font', family='serif')
plt.rcParams['lines.linewidth'] = 1
fontsize = 8
plt.rcParams['axes.labelsize'] = fontsize
plt.rcParams['xtick.labelsize'] = fontsize
plt.rcParams['axes.titlesize'] = fontsize
plt.rcParams['ytick.labelsize'] = fontsize
plt.rcParams['legend.fontsize'] = fontsize
colors = plt.rcParams['axes.prop_cycle'].by_key()['color']
configs = {
'l1': {
'norm': '$\ell_1$-norm',
'models': [
('WideResNet_28-10', 10),
('Carmon2019', 30),
('Augustin2020', 60),
],
'attacks': [
('EAD_l1_9x100', r'EAD 9$\times$100', ':', colors[0]),
('EAD_l1_9x1000', r'EAD 9$\times$1000', '-', colors[0]),
('FAB_l1_100', 'FAB $\ell_1$ 100', ':', colors[1]),
('FAB_l1_1000', 'FAB $\ell_1$ 1000', '-', colors[1]),
('ALMA_l1_100', 'ALMA 100', ':', colors[2]),
('ALMA_l1_1000', 'ALMA 1000', '-', colors[2]),
]
},
'l2': {
'norm': '$\ell_2$-norm',
'models': [
('WideResNet_28-10', 0.3),
('Carmon2019', 2.5),
('Augustin2020', 2.5),
],
'attacks': [
('DDN_100', 'DDN 100', ':', colors[0]),
('DDN_1000', 'DDN 1000', '-', colors[0]),
('FAB_l2_100', 'FAB $\ell_2$ 100', ':', colors[1]),
('FAB_l2_1000', 'FAB $\ell_2$ 1000', '-', colors[1]),
('APGD_l2', '$\mathrm{APGD}^\mathrm{T}_\mathrm{DLR}$ $\ell_2$', '-', colors[3]),
('ALMA_l2_100', 'ALMA 100', ':', colors[2]),
('ALMA_l2_1000', 'ALMA 1000', '-', colors[2]),
]
},
'ciede2000': {
'norm': 'Accumulated CIEDE2000',
'models': [
('WideResNet_28-10', 4),
('Carmon2019', 10),
('Augustin2020', 10),
],
'attacks': [
('Perc-AL_100', 'PerC-AL 100', ':', colors[0]),
('Perc-AL_1000', 'PerC-AL 1000', '-', colors[0]),
('ALMA_CIEDE2000_100', 'ALMA CIEDE2000 100', ':', colors[2]),
('ALMA_CIEDE2000_1000', 'ALMA CIEDE2000 1000', '-', colors[2]),
]
}
}
for distance, config in configs.items():
for model, xlim_high in config['models']:
fig, ax = plt.subplots(figsize=(3.2, 2.4))
for attack, legend, linestyle, color in config['attacks']:
metrics = torch.load(os.path.join(result_dir, 'metrics_{}_{}.pt'.format(model, attack)))
adv_distances = metrics['distances'][distance]
success = metrics['success']
distances, robust_accuracies = robust_accuracy_curve(distances=adv_distances, successes=success)
ax.plot(distances, robust_accuracies, label=legend, linestyle=linestyle, c=color)
ax.legend()
yticks = [0, 0.1, 0.25, 0.5, 0.75, 1]
yticklabels = [0, 10, 25, 50, 75, 100]
ax.set_yticks(yticks)
ax.set_yticklabels(yticklabels)
ax.set_xlim(0, xlim_high)
ax.set_ylim(0, 1)
ax.set_ylabel('Robust Accuracy (\%)')
ax.set_xlabel(config['norm'])
plt.grid(True, linestyle='--', c='lightgray', which='both')
fig.savefig('results/curves/attack_curves_cifar10_{}_{}.pdf'.format(distance, model), bbox_inches='tight')
configs = {
'l1': {
'norm': '$\ell_1$-norm',
'models': [
('Carmon2019', 'Carmon \emph{et al.} -- Wide ResNet 28-10', 30),
('Augustin2020', 'Augustin \emph{et al.} -- ResNet-50', 60),
],
'attacks': [
('EAD_l1_9x100', r'EAD 9$\times$100', ':', colors[0]),
('EAD_l1_9x1000', r'EAD 9$\times$1000', '-', colors[0]),
('FAB_l1_100', 'FAB $\ell_1$ 100', ':', colors[1]),
('FAB_l1_1000', 'FAB $\ell_1$ 1000', '-', colors[1]),
('ALMA_l1_100', 'ALMA 100', ':', colors[2]),
('ALMA_l1_1000', 'ALMA 1000', '-', colors[2]),
]
},
}
for distance, config in configs.items():
fig, axes = plt.subplots(1, 2, figsize=(4.8, 2.))
for i, (ax, (model, model_name, xlim_high)) in enumerate(zip(axes, config['models'])):
for attack, legend, linestyle, color in config['attacks']:
metrics = torch.load(os.path.join(result_dir, 'metrics_{}_{}.pt'.format(model, attack)))
adv_distances = metrics['distances'][distance]
success = metrics['success']
distances, robust_accuracies = robust_accuracy_curve(distances=adv_distances, successes=success)
ax.plot(distances, robust_accuracies, label=legend, linestyle=linestyle, c=color)
yticks = [0, 0.1, 0.25, 0.5, 0.75, 1]
yticklabels = [0, 10, 25, 50, 75, 100]
ax.set_title(model_name, pad=3)
ax.set_yticks(yticks)
ax.set_xlim(0, xlim_high)
ax.set_ylim(0, 1)
ax.tick_params(axis='x', pad=2)
if i == 0:
ax.legend()
ax.set_yticklabels(yticklabels)
ax.set_ylabel('Robust Accuracy (\%)', labelpad=2)
else:
ax.set_yticklabels([])
ax.set_xlabel(config['norm'], labelpad=2)
ax.grid(True, linestyle='--', c='lightgray', which='both')
fig.subplots_adjust(left=0.09, bottom=0.15, right=0.98, top=0.92, wspace=0.08, hspace=0)
fig.savefig('results/curves/attack_curves_cifar10_{}_combined.pdf'.format(distance))