-
Notifications
You must be signed in to change notification settings - Fork 4
/
KRN.py
217 lines (178 loc) · 9.36 KB
/
KRN.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
import torch
from torch import nn
import torch.nn.functional as F
from networks.deeplab_resnet import \
resnet50_locate
from networks.vgg import vgg16_locate
config_vgg = {'convert': [[128, 256, 512, 512, 512], [64, 128, 256, 512, 512]],
'deep_pool': [[512, 512, 256, 128], [512, 256, 128, 128], [True, True, True, False],
[True, True, True, False]], 'score': 128} # no convert layer, no conv6
config_resnet = {'convert': [[64, 256, 512, 1024, 2048], [128, 256, 256, 512, 512]],
'deep_pool': [[512, 512, 256, 256, 128], [512, 256, 256, 128, 128], [False, True, True, True, False],
[True, True, True, True, False]], 'score': 128}
class ConvertLayer(nn.Module):
def __init__(self, list_k):
super(ConvertLayer, self).__init__()
up = []
for i in range(len(list_k[0])):
up.append(nn.Sequential(nn.Conv2d(list_k[0][i], list_k[1][i], 1, 1, bias=False), nn.ReLU(inplace=True)))
self.convert0 = nn.ModuleList(up)
def forward(self, list_x):
resl = []
for i in range(len(list_x)):
resl.append(self.convert0[i](list_x[i]))
return resl
class DeepPoolLayer_first(nn.Module):
def __init__(self, k, k_out, need_x2,
need_fuse): # (config['deep_pool'][0][i], config['deep_pool'][1][i], config['deep_pool'][2][i], config['deep_pool'][3][i])
super(DeepPoolLayer_first, self).__init__()
self.pools_sizes = [2, 2, 2]
self.need_x2 = need_x2
self.need_fuse = need_fuse
pools, convs = [], []
for i in self.pools_sizes:
pools.append(nn.AvgPool2d(kernel_size=i, stride=i))
convs.append(nn.Conv2d(k, k, 3, 1, 1, bias=False))
self.pools = nn.ModuleList(pools)
self.convs = nn.ModuleList(convs)
self.relu = nn.ReLU()
self.conv_sum = nn.Conv2d(k, k_out, 3, 1, 1, bias=False)
if self.need_fuse:
self.conv_sum_c = nn.Conv2d(k_out, k_out, 3, 1, 1, bias=False)
def forward(self, x, x2=None): # (merge, conv2merge[k+1], infos[k])
x_size = x.size()
resl = x
y = x
for i in range(len(self.pools_sizes)):
y = self.convs[i](self.pools[i](y))
resl = torch.add(resl, F.interpolate(y, x_size[2:], mode='bilinear', align_corners=True))
resl = self.relu(resl)
if self.need_x2:
resl = F.interpolate(resl, x2.size()[2:], mode='bilinear', align_corners=True)
resl = self.conv_sum(resl)
if self.need_fuse:
resl = self.conv_sum_c(torch.add(resl, x2))
return resl
class ScoreLayer(nn.Module):
def __init__(self, k):
super(ScoreLayer, self).__init__()
self.score = nn.Conv2d(k, 1, 1, 1)
def forward(self, x, x_size=None):
x = self.score(x)
if x_size is not None:
x = F.interpolate(x, x_size[2:], mode='bilinear', align_corners=True)
return x
def extra_layer(base_model_cfg, vgg):
if base_model_cfg == 'vgg':
config = config_vgg
elif base_model_cfg == 'resnet':
config = config_resnet
convert_layers, score_layers = [], []
convert_layers = ConvertLayer(config['convert'])
score_layers = ScoreLayer(config['score'])
return vgg, convert_layers, score_layers
class KRN(nn.Module):
def __init__(self, base_model_cfg, base, convert_layers, score_layers):
super(KRN, self).__init__()
self.base_model_cfg = base_model_cfg
self.base = base # 基本网络是一样的
#self.deep_pool = nn.ModuleList(deep_pool_layers)
self.score = score_layers
if self.base_model_cfg == 'resnet':
self.convert = convert_layers
# 'deep_pool': [[512, 512, 256, 256, 128], [512, 256, 256, 128, 128], [False, True, True, True, False], [True, True, True, True, False]]
self.DeepPool_solid1 = DeepPoolLayer_first(512, 512, False, True)
self.DeepPool_solid2 = DeepPoolLayer_first(512, 256, True, True)
self.DeepPool_solid3 = DeepPoolLayer_first(256, 256, True, True)
self.DeepPool_solid4 = DeepPoolLayer_first(256, 128, True, True)
self.DeepPool_solid5 = DeepPoolLayer_first(128, 128, False, False)
self.DeepPool_contour1 = DeepPoolLayer_first(512, 512, False, True)
self.DeepPool_contour2 = DeepPoolLayer_first(512, 256, True, True)
self.DeepPool_contour3 = DeepPoolLayer_first(256, 256, True, True)
self.DeepPool_contour4 = DeepPoolLayer_first(256, 128, True, True)
self.DeepPool_contour5 = DeepPoolLayer_first(128, 128, False, False)
self.relu = nn.ReLU()
self.conv_reduce1 = nn.Conv2d(512, 128, 1, 1, 1, bias=False)
self.conv_reduce2 = nn.Conv2d(256, 128, 1, 1, 1, bias=False)
self.conv_reduce3 = nn.Conv2d(256, 128, 1, 1, 1, bias=False)
self.score_solid = ScoreLayer(128)
self.score_contour = ScoreLayer(128)
self.score_solid1 = ScoreLayer(512)
self.score_contour1 = ScoreLayer(512)
self.score_solid2 = ScoreLayer(256)
self.score_contour2 = ScoreLayer(256)
self.score_solid3 = ScoreLayer(256)
self.score_contour3 = ScoreLayer(256)
self.score_solid4 = ScoreLayer(128)
self.score_contour4 = ScoreLayer(128)
self.score_solid = ScoreLayer(128)
self.score_contour = ScoreLayer(128)
self.score_sum_out = ScoreLayer(128)
self.conv_1 = nn.Conv2d(512, 512, 3, 1, 1, bias=False)
self.conv_2 = nn.Conv2d(256, 256, 3, 1, 1, bias=False)
self.conv_3 = nn.Conv2d(256, 256, 3, 1, 1, bias=False)
self.conv_4 = nn.Conv2d(128, 128, 3, 1, 1, bias=False)
self.conv_add1 = nn.Conv2d(128, 32, 3, 1, 1, bias=False)
self.conv_add2 = nn.Conv2d(128, 32, 3, 1, 1, bias=False)
self.conv_add3 = nn.Conv2d(128, 32, 3, 1, 1, bias=False)
self.conv_add4 = nn.Conv2d(128, 32, 3, 1, 1, bias=False)
self.conv_sum_out = nn.Conv2d(128, 128, 3, 1, 1, bias=False)
def forward(self, x):
x_size = x.size()
conv2merge, infos = self.base(x) #
if self.base_model_cfg == 'resnet':
conv2merge = self.convert(conv2merge)
conv2merge = conv2merge[::-1]
merge_solid1 = self.DeepPool_solid1(conv2merge[0], conv2merge[1])
out_merge_solid1 = self.score_solid1(merge_solid1, x_size)
out_merge_solid1 = F.sigmoid(out_merge_solid1)
fea_reduce1 = self.conv_reduce1(merge_solid1)
fea_reduce1 = self.relu(fea_reduce1)
merge_solid2 = self.DeepPool_solid2(merge_solid1, conv2merge[2])
out_merge_solid2 = self.score_solid2(merge_solid2, x_size)
out_merge_solid2 = F.sigmoid(out_merge_solid2)
fea_reduce2 = self.conv_reduce2(merge_solid2)
fea_reduce2 = self.relu(fea_reduce2)
merge_solid3 = self.DeepPool_solid3(merge_solid2, conv2merge[3])
out_merge_solid3 = self.score_solid3(merge_solid3, x_size)
out_merge_solid3 = F.sigmoid(out_merge_solid3)
fea_reduce3 = self.conv_reduce3(merge_solid3)
fea_reduce3 = self.relu(fea_reduce3)
merge_solid4 = self.DeepPool_solid4(merge_solid3, conv2merge[4])
out_merge_solid4 = self.score_solid4(merge_solid4, x_size)
out_merge_solid4 = F.sigmoid(out_merge_solid4)
fea_reduce4 = merge_solid4
merge_solid5 = self.DeepPool_solid5(merge_solid4)
merge_solid = self.score_solid(merge_solid5, x_size)
merge_solid = F.sigmoid(merge_solid)
fea_reduce1 = F.interpolate(fea_reduce1, merge_solid5.size()[2:], mode='bilinear', align_corners=True)
fea_reduce2 = F.interpolate(fea_reduce2, merge_solid5.size()[2:], mode='bilinear', align_corners=True)
fea_reduce3 = F.interpolate(fea_reduce3, merge_solid5.size()[2:], mode='bilinear', align_corners=True)
fea_reduce4 = F.interpolate(fea_reduce4, merge_solid5.size()[2:], mode='bilinear', align_corners=True)
fea_add1 = torch.add(merge_solid5, fea_reduce1)
fea_add2 = torch.add(merge_solid5, fea_reduce2)
fea_add3 = torch.add(merge_solid5, fea_reduce3)
fea_add4 = torch.add(merge_solid5, fea_reduce4)
fea_add1 = self.conv_add1(fea_add1)
fea_add1 = self.relu(fea_add1)
fea_add2 = self.conv_add2(fea_add2)
fea_add2 = self.relu(fea_add2)
fea_add3 = self.conv_add3(fea_add3)
fea_add3 = self.relu(fea_add3)
fea_add4 = self.conv_add4(fea_add4)
fea_add4 = self.relu(fea_add4)
feasum_out = torch.cat((fea_add1, fea_add2, fea_add3, fea_add4), 1)
feasum_out = self.conv_sum_out(feasum_out)
feasum_out = self.score_sum_out(feasum_out, x_size)
feasum_out = F.sigmoid(feasum_out)
return feasum_out, merge_solid, out_merge_solid1, out_merge_solid2, out_merge_solid3, out_merge_solid4
def build_model(base_model_cfg='vgg'):
if base_model_cfg == 'vgg':
return KRN(base_model_cfg, *extra_layer(base_model_cfg, vgg16_locate()))
elif base_model_cfg == 'resnet':
return KRN(base_model_cfg, *extra_layer(base_model_cfg, resnet50_locate()))
def weights_init(m):
if isinstance(m, nn.Conv2d):
m.weight.data.normal_(0, 0.01)
if m.bias is not None:
m.bias.data.zero_()