Spaces:
Build error
Build error
bensonsantos
commited on
Commit
•
dfc786f
1
Parent(s):
f5b5f26
Upload model.py
Browse files
model.py
ADDED
@@ -0,0 +1,81 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch.nn as nn
|
2 |
+
import torch
|
3 |
+
from torch.nn import functional as F
|
4 |
+
from torchvision import models
|
5 |
+
|
6 |
+
class ContextualModule(nn.Module):
|
7 |
+
def __init__(self, features, out_features=512, sizes=(1, 2, 3, 6)):
|
8 |
+
super(ContextualModule, self).__init__()
|
9 |
+
self.scales = []
|
10 |
+
self.scales = nn.ModuleList([self._make_scale(features, size) for size in sizes])
|
11 |
+
self.bottleneck = nn.Conv2d(features * 2, out_features, kernel_size=1)
|
12 |
+
self.relu = nn.ReLU()
|
13 |
+
self.weight_net = nn.Conv2d(features,features,kernel_size=1)
|
14 |
+
|
15 |
+
def __make_weight(self,feature,scale_feature):
|
16 |
+
weight_feature = feature - scale_feature
|
17 |
+
return F.sigmoid(self.weight_net(weight_feature))
|
18 |
+
|
19 |
+
def _make_scale(self, features, size):
|
20 |
+
prior = nn.AdaptiveAvgPool2d(output_size=(size, size))
|
21 |
+
conv = nn.Conv2d(features, features, kernel_size=1, bias=False)
|
22 |
+
return nn.Sequential(prior, conv)
|
23 |
+
|
24 |
+
def forward(self, feats):
|
25 |
+
h, w = feats.size(2), feats.size(3)
|
26 |
+
multi_scales = [F.upsample(input=stage(feats), size=(h, w), mode='bilinear') for stage in self.scales]
|
27 |
+
weights = [self.__make_weight(feats,scale_feature) for scale_feature in multi_scales]
|
28 |
+
overall_features = [(multi_scales[0]*weights[0]+multi_scales[1]*weights[1]+multi_scales[2]*weights[2]+multi_scales[3]*weights[3])/(weights[0]+weights[1]+weights[2]+weights[3])]+ [feats]
|
29 |
+
bottle = self.bottleneck(torch.cat(overall_features, 1))
|
30 |
+
return self.relu(bottle)
|
31 |
+
|
32 |
+
class CANNet(nn.Module):
|
33 |
+
def __init__(self, load_weights=False):
|
34 |
+
super(CANNet, self).__init__()
|
35 |
+
self.seen = 0
|
36 |
+
self.context = ContextualModule(512, 512)
|
37 |
+
self.frontend_feat = [64, 64, 'M', 128, 128, 'M', 256, 256, 256, 'M', 512, 512, 512]
|
38 |
+
self.backend_feat = [512, 512, 512,256,128,64]
|
39 |
+
self.frontend = make_layers(self.frontend_feat)
|
40 |
+
self.backend = make_layers(self.backend_feat,in_channels = 512,batch_norm=True, dilation = True)
|
41 |
+
self.output_layer = nn.Conv2d(64, 1, kernel_size=1)
|
42 |
+
if not load_weights:
|
43 |
+
mod = models.vgg16(pretrained = True)
|
44 |
+
self._initialize_weights()
|
45 |
+
for i in range(len(self.frontend.state_dict().items())):
|
46 |
+
list(self.frontend.state_dict().items())[i][1].data[:] = list(mod.state_dict().items())[i][1].data[:]
|
47 |
+
|
48 |
+
def forward(self,x):
|
49 |
+
x = self.frontend(x)
|
50 |
+
x = self.context(x)
|
51 |
+
x = self.backend(x)
|
52 |
+
x = self.output_layer(x)
|
53 |
+
return x
|
54 |
+
|
55 |
+
def _initialize_weights(self):
|
56 |
+
for m in self.modules():
|
57 |
+
if isinstance(m, nn.Conv2d):
|
58 |
+
nn.init.normal_(m.weight, std=0.01)
|
59 |
+
if m.bias is not None:
|
60 |
+
nn.init.constant_(m.bias, 0)
|
61 |
+
elif isinstance(m, nn.BatchNorm2d):
|
62 |
+
nn.init.constant_(m.weight, 1)
|
63 |
+
nn.init.constant_(m.bias, 0)
|
64 |
+
|
65 |
+
def make_layers(cfg, in_channels = 3,batch_norm=False,dilation = False):
|
66 |
+
if dilation:
|
67 |
+
d_rate = 2
|
68 |
+
else:
|
69 |
+
d_rate = 1
|
70 |
+
layers = []
|
71 |
+
for v in cfg:
|
72 |
+
if v == 'M':
|
73 |
+
layers += [nn.MaxPool2d(kernel_size=2, stride=2)]
|
74 |
+
else:
|
75 |
+
conv2d = nn.Conv2d(in_channels, v, kernel_size=3, padding=d_rate,dilation = d_rate)
|
76 |
+
if batch_norm:
|
77 |
+
layers += [conv2d, nn.BatchNorm2d(v), nn.ReLU(inplace=True)]
|
78 |
+
else:
|
79 |
+
layers += [conv2d, nn.ReLU(inplace=True)]
|
80 |
+
in_channels = v
|
81 |
+
return nn.Sequential(*layers)
|