Coverage for src/deepdraw/models/driu.py: 75%

1# SPDX-FileCopyrightText: Copyright © 2023 Idiap Research Institute <contact@idiap.ch> 

2# 

3# SPDX-License-Identifier: GPL-3.0-or-later 

4 

5from collections import OrderedDict 

6 

7import torch 

8import torch.nn 

9 

10from .backbones.vgg import vgg16_for_segmentation 

11from .make_layers import UpsampleCropBlock, conv_with_kaiming_uniform 

12 

13 

14class ConcatFuseBlock(torch.nn.Module): 

15 """Takes in four feature maps with 16 channels each, concatenates them and 

16 applies a 1x1 convolution with 1 output channel.""" 

17 

18 def __init__(self): 

19 super().__init__() 

20 self.conv = conv_with_kaiming_uniform(4 * 16, 1, 1, 1, 0) 

21 

22 def forward(self, x1, x2, x3, x4): 

23 x_cat = torch.cat([x1, x2, x3, x4], dim=1) 

24 x = self.conv(x_cat) 

25 return x 

26 

27 

28class DRIU(torch.nn.Module): 

29 """DRIU head module. 

30 

31 Based on paper by [MANINIS-2016]_. 

32 

33 Parameters 

34 ---------- 

35 in_channels_list : list 

36 number of channels for each feature map that is returned from backbone 

37 """ 

38 

39 def __init__(self, in_channels_list=None): 

40 super().__init__() 

41 ( 

42 in_conv_1_2_16, 

43 in_upsample2, 

44 in_upsample_4, 

45 in_upsample_8, 

46 ) = in_channels_list 

47 

48 self.conv1_2_16 = torch.nn.Conv2d(in_conv_1_2_16, 16, 3, 1, 1) 

49 # Upsample layers 

50 self.upsample2 = UpsampleCropBlock(in_upsample2, 16, 4, 2, 0) 

51 self.upsample4 = UpsampleCropBlock(in_upsample_4, 16, 8, 4, 0) 

52 self.upsample8 = UpsampleCropBlock(in_upsample_8, 16, 16, 8, 0) 

53 

54 # Concat and Fuse 

55 self.concatfuse = ConcatFuseBlock() 

56 

57 def forward(self, x): 

58 """ 

59 

60 Parameters 

61 ---------- 

62 

63 x : list 

64 list of tensors as returned from the backbone network. First 

65 element: height and width of input image. Remaining elements: 

66 feature maps for each feature level. 

67 

68 Returns 

69 ------- 

70 

71 tensor : :py:class:`torch.Tensor` 

72 

73 """ 

74 hw = x[0] 

75 conv1_2_16 = self.conv1_2_16(x[1]) # conv1_2_16 

76 upsample2 = self.upsample2(x[2], hw) # side-multi2-up 

77 upsample4 = self.upsample4(x[3], hw) # side-multi3-up 

78 upsample8 = self.upsample8(x[4], hw) # side-multi4-up 

79 out = self.concatfuse(conv1_2_16, upsample2, upsample4, upsample8) 

80 return out 

81 

82 

83def driu(pretrained_backbone=True, progress=True): 

84 """Builds DRIU for vessel segmentation by adding backbone and head 

85 together. 

86 

87 Parameters 

88 ---------- 

89 

90 pretrained_backbone : :py:class:`bool`, Optional 

91 If set to ``True``, then loads a pre-trained version of the backbone 

92 (not the head) for the DRIU network using VGG-16 trained for ImageNet 

93 classification. 

94 

95 progress : :py:class:`bool`, Optional 

96 If set to ``True``, and you decided to use a ``pretrained_backbone``, 

97 then, shows a progress bar of the backbone model downloading if 

98 download is necesssary. 

99 

100 

101 Returns 

102 ------- 

103 

104 module : :py:class:`torch.nn.Module` 

105 Network model for DRIU (vessel segmentation) 

106 """ 

107 

108 backbone = vgg16_for_segmentation( 

109 pretrained=pretrained_backbone, 

110 progress=progress, 

111 return_features=[3, 8, 14, 22], 

112 ) 

113 head = DRIU([64, 128, 256, 512]) 

114 

115 order = [("backbone", backbone), ("head", head)] 

116 if pretrained_backbone: 

117 from .normalizer import TorchVisionNormalizer 

118 

119 order = [("normalizer", TorchVisionNormalizer())] + order 

120 

121 model = torch.nn.Sequential(OrderedDict(order)) 

122 model.name = "driu" 

123 return model