File size: 4,372 Bytes
165f2cc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import torch
import torch.nn as nn
import torch.nn.functional as F
import timm
from typing import Optional

class DecoderBlock(nn.Module):
  def __init__(self, in_channels: int, skip_channels: int, out_channels: int):
    super(DecoderBlock, self).__init__()
    self.up = nn.ConvTranspose2d(in_channels, out_channels, kernel_size=2, stride=2)
    self.conv1 = nn.Conv2d(out_channels + skip_channels, out_channels, kernel_size=3, padding=1)
    self.bn1 = nn.BatchNorm2d(out_channels)
    self.relu1 = nn.ReLU(inplace=True)
    self.conv2 = nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1)
    self.bn2 = nn.BatchNorm2d(out_channels)
    self.relu2 = nn.ReLU(inplace=True)

  def forward(self, x: torch.Tensor, skip: Optional[torch.Tensor]) -> torch.Tensor:
    x = self.up(x)
    if skip is not None:
      if x.size() != skip.size():
        x = F.interpolate(x, size=skip.shape[2:], mode='bilinear', align_corners=True)
      x = torch.cat([x, skip], dim=1)
    x = self.conv1(x)
    x = self.bn1(x)
    x = self.relu1(x)
    x = self.conv2(x)
    x = self.bn2(x)
    x = self.relu2(x)
    return x

class UNetTimmWithClassification(nn.Module):
  def __init__(self,
      encoder_name: str = 'resnet50',
      encoder_weights: Optional[str] = 'imagenet',
      num_classes_seg: int = 1,
      num_classes_cls: int = 9
  ):

      super(UNetTimmWithClassification, self).__init__()

      self.encoder = timm.create_model(
          encoder_name,
          pretrained=(encoder_weights == 'imagenet'),
          features_only=True,
          in_chans=3
      )

      encoder_channels = self.encoder.feature_info.channels()
      decoder_channels = [2048, 1024, 512, 256]
      # decoder_channels = [512, 256, 128, 64]

      self.decoder4 = DecoderBlock(
          in_channels=decoder_channels[0],
          skip_channels=encoder_channels[3],
          out_channels=decoder_channels[1]
      )
      self.decoder3 = DecoderBlock(
          in_channels=decoder_channels[1],
          skip_channels=encoder_channels[2],
          out_channels=decoder_channels[2]
      )
      self.decoder2 = DecoderBlock(
          in_channels=decoder_channels[2],
          skip_channels=encoder_channels[1],
          out_channels=decoder_channels[3]
      )
      self.decoder1 = DecoderBlock(
          in_channels=decoder_channels[3],
          skip_channels=encoder_channels[0],
          out_channels=decoder_channels[3]
      )
      self.final_up = nn.ConvTranspose2d(
          in_channels=decoder_channels[-1],
          out_channels=32,
          kernel_size=2,
          stride=2
      )
      self.final_conv_seg = nn.Conv2d(
          in_channels=32,
          out_channels=num_classes_seg,
          kernel_size=1
      )

      #Cls head
      self.classification_head = nn.Sequential(
          nn.AdaptiveAvgPool2d(1),
          nn.Flatten(),
          nn.Linear(2048, 512),
          # nn.Linear(encoder_channels[-1], 512),
          nn.Dropout(0.2),
          nn.Linear(512, 512),
          nn.ReLU(),
          nn.Linear(512, num_classes_cls)
      )

      if num_classes_cls > 1:
          self.classification_activation = nn.Softmax(dim=1)
      elif num_classes_cls == 1:
          self.classification_activation = nn.Sigmoid()
      else:
          self.classification_activation = None

      if self.classification_activation is not None:
          self.classification_head.add_module("activation", self.classification_activation)

      #Xavier weight initialize
      if encoder_weights == 'xavier':
        self.apply(self.xavier_init_weights)

  def xavier_init_weights(self, m):
    if isinstance(m, nn.Conv2d) or isinstance(m, nn.ConvTranspose2d):
      nn.init.xavier_uniform_(m.weight)
      if m.bias is not None:
        nn.init.zeros_(m.bias)
    elif isinstance(m, nn.Linear):
      nn.init.xavier_uniform_(m.weight)
      if m.bias is not None:
        nn.init.zeros_(m.bias)
    elif isinstance(m, nn.BatchNorm2d):
      nn.init.ones_(m.weight)
      nn.init.zeros_(m.bias)

  def forward(self, x: torch.Tensor) -> tuple:
    features = self.encoder(x)
    C0, C1, C2, C3, C4 = features
    cls = self.classification_head(C4)
    D4 = self.decoder4(C4, C3)
    D3 = self.decoder3(D4, C2)
    D2 = self.decoder2(D3, C1)
    D1 = self.decoder1(D2, C0)
    x = self.final_up(D1)
    seg = self.final_conv_seg(x)
    return seg, cls