cosnet/cosnet_modules.py at main · deeprightai/cosnet · GitHub

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
import torch
from torch import nn
from timm.models.layers import DropPath
import torch.nn.functional as F
from timm.models.layers import trunc_normal_


class LayerNorm(nn.Module):
    def __init__(self, normalized_shape, eps=1e-6, data_format="channels_last"):
        super().__init__()
        self.weight = nn.Parameter(torch.ones(normalized_shape))
        self.bias = nn.Parameter(torch.zeros(normalized_shape))
        self.eps = eps
        self.data_format = data_format
        if self.data_format not in ["channels_last", "channels_first"]:
            raise NotImplementedError
        self.normalized_shape = (normalized_shape,)

    def forward(self, x):
        if self.data_format == "channels_last":
            return F.layer_norm(x, self.normalized_shape, self.weight, self.bias, self.eps)
        elif self.data_format == "channels_first":
            u = x.mean(1, keepdim=True)
            s = (x - u).pow(2).mean(1, keepdim=True)
            x = (x - u) / torch.sqrt(s + self.eps)
            x = self.weight[:, None, None] * x + self.bias[:, None, None]
            return x


class MCFS(nn.Module):
    def __init__(self, dim, s_kernel_size=3):
        super().__init__()

        self.proj_1 = nn.Conv2d(dim, dim, 1, padding=0)
        self.proj_2 = nn.Conv2d(dim*2, dim, 1, padding=0)
        self.norm_proj = LayerNorm(dim, eps=1e-6, data_format="channels_first")

        # multiscale spatial context layers
        self.s_ctx_1 = nn.Conv2d(dim, dim, kernel_size=s_kernel_size, padding=s_kernel_size//2, groups=dim//4)
        self.s_ctx_2 = nn.Conv2d(dim, dim, kernel_size=s_kernel_size, dilation=2, padding=(s_kernel_size//2)*2, groups=dim//4)
        self.norm_s = LayerNorm(dim, eps=1e-6, data_format="channels_first")

        # sharpening module layers
        self.h_ctx = nn.Conv2d(dim, dim, kernel_size=5, padding=2, bias=False, groups=dim)
        self.norm_h = LayerNorm(dim, eps=1e-6, data_format="channels_first")

        self.act = nn.GELU()


    def forward(self, x):
        x = self.norm_proj(self.act(self.proj_1(x)))
        B, C, H, W = x.shape

        # extract multi-scale contextual features
        sx1 = self.act(self.s_ctx_1(x))
        sx2 = self.act(self.s_ctx_2(x))

        sx = self.norm_s(sx1 + sx2)

        # feature enhancement using learnable sharpening factors
        # implementation of sharpening module
        hx = self.act(self.h_ctx(x))
        hx_t = x - hx.mean(dim=1).unsqueeze(1)
        hx_t = torch.softmax(hx.mean(dim=[-2,-1]).unsqueeze(-1).unsqueeze(-1), dim=1) * hx_t
        hx = self.norm_h(hx + hx_t)

        # combine the multiscale contetxual features with the sharpened features
        x = self.act(self.proj_2(torch.cat([sx, hx], dim=1)))

        return x


class FSB(nn.Module):
    """
    Feature Sharpening Block:
    It is the core block of the COSNet encoder/backbone,
    utilized to extract semantically rich features for segementation task in cluttered background.
    -----------------------------------------------
    dim:           Input channel dimension
    s_kernel_size: Kernel size for spatial context layers
    expan_ratio:   Expansion ratio used for channels in MLP
    ------------------------------------------------

    """
    def __init__(self, dim, s_kernel_size=3, drop_path=0.1, layer_scale_init_value=1e-6, expan_ratio=4):
        super().__init__()

        self.conv_dw = nn.Conv2d(dim, dim, kernel_size=3, padding=1, groups=dim)
        self.norm_dw = LayerNorm(dim, eps=1e-6, data_format="channels_first")

        self.layer_norm_1 = LayerNorm(dim, eps=1e-6, data_format="channels_first")
        self.layer_norm_2 = LayerNorm(dim, eps=1e-6, data_format="channels_first")

        self.mlp = MLP(dim=dim, mlp_ratio=expan_ratio)
        self.attn = MCFS(dim, s_kernel_size=s_kernel_size)

        self.drop_path_1 = DropPath(drop_path) if drop_path > 0. else nn.Identity()
        self.drop_path_2 = DropPath(drop_path) if drop_path > 0. else nn.Identity()

        self.act = nn.GELU()


    def forward(self, x):
        B, C, H, W = x.shape

        x = x + self.norm_dw(self.act(self.conv_dw(x)))

        x_copy = x
        x = self.layer_norm_1(x_copy)
        x = self.drop_path_1(self.attn(x))
        out = x + x_copy

        x = self.layer_norm_2(out)
        x = self.drop_path_2(self.mlp(x))
        out = out + x

        return out


class MLP(nn.Module):
    def __init__(self, dim, mlp_ratio=4):
        super().__init__()

        self.fc_1 = nn.Conv2d(dim, dim * mlp_ratio, 1)
        self.pos = nn.Conv2d(dim * mlp_ratio, dim * mlp_ratio, 3, padding=1, groups=dim * mlp_ratio)
        self.fc_2 = nn.Conv2d(dim * mlp_ratio, dim, 1)
        self.act = nn.GELU()

    def forward(self, x):
        B, C, H, W = x.shape

        x = self.fc_1(x)
        x = self.act(x)
        x = x + self.act(self.pos(x))
        x = self.fc_2(x)

        return x


class BEM(nn.Module):
    def __init__(self, dim):
        super().__init__()
        self.conv = nn.Conv2d(dim*2, dim, kernel_size=3, stride=1, padding=1)
        self.norm = LayerNorm(dim, eps=1e-6, data_format="channels_first")
        self.act = nn.GELU()
        self.pool = nn.MaxPool2d(kernel_size=4, stride=4)

    def forward(self, x):
        dx = self.pool(x)
        ex = torch.nn.functional.interpolate(dx, size=x.shape[2:], mode='bilinear') - x
        x = torch.cat([ex,x], dim=1)
        x = self.conv(x)
        x = self.act(x)
        x = self.norm(x)

        return x