-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathmodel.py
174 lines (139 loc) · 6.28 KB
/
model.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
import numpy as np
import torch
import torch.nn as nn
import torchvision.models as models
from ResNet import ResNet50
from torch.nn import functional as F
from modules import norm_layer, ChannelCompression, upsample, PredictLayer
from vit_seg_modeling import Transformer, CONFIGS
CHANNELS = [64, 256, 512, 1024]
class Net(nn.Module):
def __init__(self, config):
super(Net, self).__init__()
self.config = config
# Backbone model
self.resnet = ResNet50('rgb')
self.resnet_depth = ResNet50('rgbd')
self.fuse_backbone = Transformer(config=CONFIGS['R50-ViT-B_16'], img_size=config.trainsize, vis=False)
# fusion
from modules import CrossAttentionFusionPool as FusionBlock
from modules import Decoder as DecoderBlock
fusions = []
for i in range(4):
fusions.append(FusionBlock(CHANNELS[i], config.dilation, config.kernel))
self.fusion = nn.ModuleList(fusions)
predicts, decoders = [], []
for i in range(config.decoder_num):
decoders.append(DecoderBlock(64))
predicts.append(PredictLayer(64, config.trainsize))
self.decoders = nn.ModuleList(decoders)
self.predicts = nn.ModuleList(predicts)
self.rgb_compress = nn.Sequential(
nn.Conv2d(2048, 1024, 1, bias=False),
norm_layer(1024),
nn.ReLU(inplace=True)
)
self.depth_compress = nn.Sequential(
nn.Conv2d(2048, 1024, 1, bias=False),
norm_layer(1024),
nn.ReLU(inplace=True)
)
self.compress1 = ChannelCompression(256, 64)
self.compress2 = ChannelCompression(512, 64) #2
self.compress3 = ChannelCompression(768, 64)
# Prediction
self.high_predict = PredictLayer(64, config.trainsize)
self.middle_predict = PredictLayer(64, config.trainsize)
if self.training:
self.initialize_weights()
def forward(self, x, x_depth):
# stage 0
x = self.resnet.conv1(x)
x = self.resnet.bn1(x)
x = self.resnet.relu(x)
x = self.resnet.maxpool(x)
x_depth = self.resnet_depth.conv1(x_depth)
x_depth = self.resnet_depth.bn1(x_depth)
x_depth = self.resnet_depth.relu(x_depth)
x_depth = self.resnet_depth.maxpool(x_depth)
x, x_depth, x_fused = self.fusion[0](x, x_depth)
# stage 1
x1 = self.resnet.layer1(x)
x1_depth = self.resnet_depth.layer1(x_depth)
x1_fused = self.fuse_backbone.embeddings.hybrid_model.body[0](x_fused)
x1, x1_depth, x1_fused_add = self.fusion[1](x1, x1_depth)
x1_fused = x1_fused_add + x1_fused
# stage 2
x2 = self.resnet.layer2(x1)
x2_depth = self.resnet_depth.layer2(x1_depth)
x2_fused = self.fuse_backbone.embeddings.hybrid_model.body[1](x1_fused)
x2, x2_depth, x2_fused_add = self.fusion[2](x2, x2_depth)
x2_fused = x2_fused_add + x2_fused
# stage 3
x3 = self.resnet.layer4(self.resnet.layer3(x2))
x3_depth = self.resnet_depth.layer4(self.resnet_depth.layer3(x2_depth))
x3_fused = self.fuse_backbone.embeddings.hybrid_model.body[2](x2_fused)
x3 = self.rgb_compress(x3)
x3_depth = self.depth_compress(x3_depth)
x3, x3_depth, x3_fused_add = self.fusion[3](x3, x3_depth)
x3_fused = x3_fused_add + x3_fused
# delete intermediate features
del x, x_depth, x_fused, x1, x1_depth, x2, x2_depth, x3, x3_depth
del x1_fused_add, x2_fused_add, x3_fused_add
# transformer part
x3_fused = self.fuse_backbone.embeddings.patch_embeddings(x3_fused)
x3_fused = x3_fused.flatten(2)
x3_fused = x3_fused.transpose(-1, -2)
x3_fused = x3_fused + self.fuse_backbone.embeddings.position_embeddings
x3_fused = self.fuse_backbone.embeddings.dropout(x3_fused)
x3_fused, _ = self.fuse_backbone.encoder(x3_fused)
B, n_patch, hidden = x3_fused.size()
h, w = int(np.sqrt(n_patch)), int(np.sqrt(n_patch))
x3_fused = x3_fused.permute(0, 2, 1)
x3_fused = x3_fused.contiguous().view(B, hidden, h, w)
# channel compression
x1_fused = self.compress1(x1_fused)
x2_fused = self.compress2(x2_fused)
x3_fused = self.compress3(x3_fused)
# decoder
decoder_output = []
for i in range(self.config.decoder_num):
x3_fused, x2_fused, x1_fused = self.decoders[i](x3_fused, x2_fused, x1_fused)
decoder_output.append(self.predicts[i](x1_fused))
# prediction
middle_sal = self.middle_predict(x2_fused)
high_sal = self.high_predict(x3_fused)
return high_sal, middle_sal, decoder_output
# initialize the weights
def initialize_weights(self):
res50 = models.resnet50(pretrained=True)
pretrained_dict = res50.state_dict()
all_params = {}
for k, v in self.resnet.state_dict().items():
if k in pretrained_dict.keys():
v = pretrained_dict[k]
all_params[k] = v
assert len(all_params.keys()) == len(self.resnet.state_dict().keys())
self.resnet.load_state_dict(all_params)
all_params = {}
for k, v in self.resnet_depth.state_dict().items():
if k == 'conv1.weight':
all_params[k] = torch.nn.init.normal_(v, mean=0, std=1)
elif k in pretrained_dict.keys():
v = pretrained_dict[k]
all_params[k] = v
assert len(all_params.keys()) == len(self.resnet_depth.state_dict().keys())
self.resnet_depth.load_state_dict(all_params)
self.fuse_backbone.load_from(np.load('./models/imagenet21k_R50+ViT-B_16.npz'))
if __name__ == '__main__':
model = Net().cuda()
img = torch.randn((3, 3, 256, 256)).cuda()
with torch.no_grad():
res1 = model.resnet.maxpool(model.resnet.relu(model.resnet.conv1(img))) # (3, 64, 64, 64)
res2 = model.resnet.layer1(res1) # (3, 256, 64, 64)
res3 = model.resnet.layer2(res2) # （3，512，32，32）
res4 = model.resnet.layer3_1(res3) # （3，1024，16，16）
res5 = model.resnet.layer4_1(res4) # （3，2048，8，8）
features = [res1, res2, res3, res4, res5]
for i in range(5):
print(features[i].shape)