-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathAverageAnalysis.py
242 lines (215 loc) · 11.5 KB
/
AverageAnalysis.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
import os
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import plotly
plotly.offline.init_notebook_mode(connected=True)
import plotly.graph_objs as go
def load_sims(model_name='',path='./results', show=True):
sim_path = os.path.join(path, model_name, 'sims')
layer_names = os.listdir(sim_path)
all_layer_sims = []
for layer_name in layer_names:
layer_sims = []
layer_path = os.path.join(sim_path, layer_name)
for sim_name in os.listdir(layer_path):
layer_sims.append(np.load(os.path.join(layer_path, sim_name), allow_pickle=True).squeeze())
all_layer_sims.append(np.array(layer_sims))
if show:
print('Load all layer similarities(Layer-Class-Kernel): ', model_name, len(all_layer_sims))
return all_layer_sims
def load_vis_results(model_name='',path='./results', show=True):
vis_path = os.path.join(path, model_name, 'vis_results')
layer_names = os.listdir(vis_path)
all_layer_imgs = []
for layer_name in layer_names:
layer_vis = []
layer_path = os.path.join(vis_path, layer_name)
for cls_name in os.listdir(layer_path):
cls_path = os.path.join(layer_path, cls_name)
cls_vis = []
for img_name in os.listdir(cls_path):
cls_vis.append(plt.imread(os.path.join(cls_path, img_name)))
layer_vis.append(np.array(cls_vis))
all_layer_imgs.append(np.array(layer_vis))
if show:
print('Load all layer vis results(Layer-Class-Kernel): ', model_name, len(all_layer_imgs))
return all_layer_imgs
def load_sims_and_vis_results(model_name='',path='./results'):
all_layer_sims = load_sims(model_name=model_name, path=path)
all_layer_vis_results = load_vis_results(model_name=model_name, path=path)
return all_layer_sims, all_layer_vis_results
def draw_sim_one_layer(sim_data_one_layer, model_name=''):
data = []
for i in range(sim_data_one_layer.shape[0]):
data.append(go.Scatter(x=np.arange(sim_data_one_layer[i].shape[0]), y=sim_data_one_layer[i], name='cls_'+str(i),mode='markers'))
data.append(go.Scatter(x=np.arange(sim_data_one_layer[i].shape[0]), y=sim_data_one_layer.mean(axis=0), name='均值',mode='lines+markers'))
fig = go.Figure(data)
fig.update_layout(
title=model_name+'卷积核相似值',
xaxis_title='卷积核编号',
yaxis_title='相似值'
)
fig.show()
def draw_sim_all_layers(sim_data, model_name=''):
for i in range(len(sim_data)):
draw_sim_one_layer(sim_data_one_layer=sim_data[i], model_name=model_name+'层'+str(i))
def draw_sim_mean_all_class(sim_data, model_name=''):
layer_num = len(sim_data)
class_num = sim_data[0].shape[0]
cls_sim_data = []
for layer in range(layer_num):
cls_sim_data.append(sim_data[layer].mean(axis=1))
cls_sim_data = np.array(cls_sim_data).T
data = []
for i in range(class_num):
data.append(go.Scatter(x=np.arange(layer_num), y=cls_sim_data[i], name='cls_'+str(i),mode='lines+markers'))
data.append(go.Scatter(x=np.arange(layer_num), y=cls_sim_data.mean(axis=0), name='均值',mode='lines+markers'))
fig = go.Figure(data)
fig.update_layout(
title=model_name+'层相似值均值',
xaxis_title='卷积层编号',
yaxis_title='相似值'
)
fig.show()
def draw_sim_std_all_class(sim_data, model_name=''):
layer_num = len(sim_data)
class_num = sim_data[0].shape[0]
cls_sim_data = []
for layer in range(layer_num):
cls_sim_data.append(sim_data[layer].std(axis=1))
cls_sim_data = np.array(cls_sim_data).T
data = []
for i in range(class_num):
data.append(go.Scatter(x=np.arange(layer_num), y=cls_sim_data[i], name='cls_'+str(i),mode='lines+markers'))
data.append(go.Scatter(x=np.arange(layer_num), y=cls_sim_data.mean(axis=0), name='均值',mode='lines+markers'))
fig = go.Figure(data)
fig.update_layout(
title=model_name+'层相似值标准差',
xaxis_title='卷积层编号',
yaxis_title='标准差'
)
fig.show()
def draw_vis_results_one_layer(sim_one_layer, layer_path, layer_num, model_name='', index=None, pic_one_row=6, head_and_tail=True):
kernel_num = sim_one_layer.shape[1]
if index is None:
index = []
for i in range(sim_one_layer.shape[0]):
index.append(list(range(kernel_num)))
for cls, cls_name in enumerate(os.listdir(layer_path)):
cls_path = os.path.join(layer_path, cls_name)
if not head_and_tail:
lines = kernel_num//pic_one_row
if kernel_num % pic_one_row != 0:
lines+=1
figure, axes = plt.subplots(lines, pic_one_row, figsize=(16, 3*lines), tight_layout=True, clear=True)
for i in range(lines):
for j in range(pic_one_row):
if i*pic_one_row+j==kernel_num:
break
axe = axes[i, j]
pic_path = os.path.join(cls_path, str(index[cls][i*pic_one_row+j]).zfill(3)+'.jpg')
img = plt.imread(pic_path)
axe.imshow(img)
round_sim = np.round(sim_one_layer[cls][index[cls][i*pic_one_row+j]],3)
axe.set_title('Kernel: '+str(index[cls][i*pic_one_row+j])+'\nSim: '+str(round_sim))
figure.suptitle('Model:'+model_name+' Layer:'+str(layer_num)+' Class:'+cls_name,fontsize=10)
figure.tight_layout()
else:
lines=pic_one_row//8
figure, axes = plt.subplots(lines, 8, figsize=(16, 3*lines), tight_layout=True, clear=True)
for i in range(lines):
for j in range(8):
axe = axes[i,j]
pic_path = os.path.join(cls_path, str(index[cls][i*pic_one_row+j]).zfill(3)+'.jpg')
img = plt.imread(pic_path)
axe.imshow(img)
round_sim = np.round(sim_one_layer[cls][index[cls][i*pic_one_row+j]],3)
axe.set_title('Kernel: '+str(index[cls][i*pic_one_row+j])+'\nSim: '+str(round_sim))
figure.suptitle('Model:'+model_name+' Layer:'+str(layer_num)+' Class:'+cls_name,fontsize=15)
figure, axes = plt.subplots(lines, 8, figsize=(16, 3*lines), tight_layout=True, clear=True)
for i in range(lines):
for j in range(8):
axe = axes[i,j]
pic_path = os.path.join(cls_path, str(index[cls][-(i*pic_one_row+j+1)]).zfill(3)+'.jpg')
img = plt.imread(pic_path)
axe.imshow(img)
round_sim = np.round(sim_one_layer[cls][index[cls][-(i*pic_one_row+j+1)]],3)
axe.set_title('Kernel: '+str(index[cls][-(i*pic_one_row+j+1)])+'\nSim: '+str(round_sim))
def draw_sorted_vis_results_one_layer(sim_one_layer, layer_path, layer_num, model_name='', pic_one_row=6, head_and_tail=True):
argsort_sims = sim_one_layer.argsort(axis=1)
draw_vis_results_one_layer(sim_one_layer=sim_one_layer,
layer_path=layer_path,
layer_num=layer_num,
model_name=model_name,
index=argsort_sims,
pic_one_row=pic_one_row,
head_and_tail=head_and_tail)
def draw_vis_results_one_cls(sim_one_layer, layer_path, layer_num, cls, model_name='', index=None, pic_one_row=6, head_and_tail=True):
kernel_num = sim_one_layer.shape[1]
if index is None:
index = []
for i in range(sim_one_layer.shape[0]):
index.append(list(range(kernel_num)))
cls_name = os.listdir(layer_path)[cls]
cls_path = os.path.join(layer_path, cls_name)
lines=pic_one_row//8
figure, axes = plt.subplots(lines, 8, figsize=(16, 3*lines), tight_layout=True, clear=True)
for i in range(lines):
for j in range(8):
axe = axes[i,j]
pic_path = os.path.join(cls_path, str(index[cls][i*pic_one_row+j]).zfill(3)+'.jpg')
img = plt.imread(pic_path)
axe.imshow(img)
round_sim = np.round(sim_one_layer[cls][index[cls][i*pic_one_row+j]],3)
axe.set_title('Kernel: '+str(index[cls][i*pic_one_row+j])+'\nSim: '+str(round_sim))
figure.suptitle('Model:'+model_name+' Layer:'+str(layer_num)+' Class:'+cls_name,fontsize=15)
figure, axes = plt.subplots(lines, 8, figsize=(16, 3*lines), tight_layout=True, clear=True)
for i in range(lines):
for j in range(8):
axe = axes[i,j]
pic_path = os.path.join(cls_path, str(index[cls][-(i*pic_one_row+j+1)]).zfill(3)+'.jpg')
img = plt.imread(pic_path)
axe.imshow(img)
round_sim = np.round(sim_one_layer[cls][index[cls][-(i*pic_one_row+j+1)]],3)
axe.set_title('Kernel: '+str(index[cls][-(i*pic_one_row+j+1)])+'\nSim: '+str(round_sim))
def draw_sorted_vis_results_one_cls(sim_one_layer, layer_path, layer_num, cls, model_name='', pic_one_row=6, head_and_tail=True):
argsort_sims = sim_one_layer.argsort(axis=1)
draw_vis_results_one_cls(sim_one_layer=sim_one_layer,
layer_path=layer_path,
layer_num=layer_num,
cls=cls,
model_name=model_name,
index=argsort_sims,
pic_one_row=pic_one_row,
head_and_tail=head_and_tail)
def draw_vis_results_one_cls_all_layers(vis_results, sims, cls, path='./results', model_name='', pic_one_row=6, sort=True, head_and_tail=True):
path = os.path.join(path, model_name, 'vis_results')
for i, layer_name in enumerate(os.listdir(path)):
layer_path = os.path.join(path, layer_name)
draw_sorted_vis_results_one_cls(sim_one_layer=sims[i],
layer_path=layer_path,
layer_num=i,
cls=cls,
model_name=model_name,
pic_one_row=pic_one_row,
head_and_tail=head_and_tail)
def draw_vis_results_all_layers(vis_results, sims, path='./results', model_name='', pic_one_row=6, sort=True, head_and_tail=True):
path = os.path.join(path, model_name, 'vis_results')
for i, layer_name in enumerate(os.listdir(path)):
layer_path = os.path.join(path, layer_name)
if sort:
draw_sorted_vis_results_one_layer(sim_one_layer=sims[i],
layer_path=layer_path,
layer_num=i,
model_name=model_name,
pic_one_row=pic_one_row,
head_and_tail=head_and_tail)
else:
draw_vis_results_one_layer(flat_sim_layer=sims[i],
layer_path=layer_path,
layer_num=i,
model_name=model_name,
index=None,
pic_one_row=pic_one_row,
head_and_tail=head_and_tail)