workprogress/MRFNet.py at main · dty00/workprogress · 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
import torch
import torch.nn as nn
import torch.nn.functional as F

class MRFNet(nn.Module):
    def __init__(self, EncodingDepth = 8, initial_num_layers = 2048):
        super(MRFNet, self).__init__()

        self.EncodingDepth = EncodingDepth

        self.init1 = CConv1d_BN_RELU(1000, initial_num_layers, 1, 0)

        self.midLayers = []
        temp = list(range(1, EncodingDepth + 1))
        for encodingLayer in temp:
            inl = initial_num_layers // (2 ** (encodingLayer - 1))
            outl = initial_num_layers // (2 ** encodingLayer)
            self.midLayers.append(CConv1d_BN_RELU(inl, outl, 1, pad = 0))
        self.midLayers = nn.ModuleList(self.midLayers)

        self.final = nn.Linear(outl * 2, 1)

    def forward(self, x_r, x_i):
        #print(x_r.shape)
        x_r = x_r.permute(0, 2, 1)  ## nb * 1000 * 1
        x_i = x_i.permute(0, 2, 1)

        x_r, x_i  = self.init1(x_r, x_i)

        temp = list(range(1, self.EncodingDepth + 1))
        for encodingLayer in temp:
            temp_conv = self.midLayers[encodingLayer - 1]
            x_r, x_i = temp_conv(x_r, x_i)

        x_r = x_r.permute(0, 2, 1)
        x_i = x_i.permute(0, 2, 1)

        x = torch.cat([x_r, x_i], dim = -1)

        #print(x.size())
        x = self.final(x)
        t1 = x[0]
        t2 = x[1]
        b0 = x[2]
        return t1, t2, b0


class Basic_block(nn.Module):
    def __init__(self, num_in, num_out):
        super(Basic_block, self).__init__()
        self.cconv1 = CConv1d_BN_RELU(num_in, num_out)
        self.cconv2 = CConv1d_BN_RELU(num_out, num_out)

    def forward(self, x_r, x_i):
        INPUT_r = x_r
        INPUT_i = x_i
        x_r, x_i = self.cconv1(x_r, x_i)
        x_r = x_r + INPUT_r
        x_i = x_i + INPUT_i
        x_r, x_i = self.cconv2(x_r, x_i)
        return x_r, x_i

## complex convolution;
class CConv1d_BN_RELU(nn.Module):
    def __init__(self, num_in, num_out, ks = 3, pad = 1):
        super(CConv1d_BN_RELU, self).__init__()
        self.conv_r = nn.Conv1d(num_in, num_out, ks, padding= pad)
        self.conv_i = nn.Conv1d(num_in, num_out, ks, padding= pad)
        # self.bn_r = nn.BatchNorm1d(num_out)
        # self.bn_i = nn.BatchNorm1d(num_out)
        self.relu_r = nn.ReLU(inplace = True)
        self.relu_i = nn.ReLU(inplace = True)

    def forward(self, x_r, x_i):
        x_rr = self.conv_r(x_r)
        x_ri = self.conv_i(x_r)
        x_ir = self.conv_r(x_i)
        x_ii = self.conv_i(x_i)
        x_r = x_rr - x_ii
        x_i = x_ri + x_ir
        # x_r = self.bn_r(x_r)
        # x_i = self.bn_i(x_i)
        x_r = self.relu_r(x_r)
        x_i = self.relu_i(x_i)
        return x_r, x_i


## complex convolution;
class CConv1d(nn.Module):
    def __init__(self, num_in, num_out, ks = 1, pad = 0, bs = True):
        super(CConv1d, self).__init__()
        self.conv_r = nn.Conv1d(num_in, num_out, ks, bias = bs, padding= pad)
        self.conv_i = nn.Conv1d(num_in, num_out, ks, bias = bs, padding= pad)

    def forward(self, x_r, x_i):
        x_rr = self.conv_r(x_r)
        x_ri = self.conv_i(x_r)
        x_ir = self.conv_r(x_i)
        x_ii = self.conv_i(x_i)
        x_r = x_rr - x_ii
        x_i = x_ri + x_ir

        return x_r, x_i

def weights_init(m):
    if isinstance(m, nn.Conv1d):
        nn.init.normal_(m.weight, mean=0.0, std=1e-2)
        nn.init.zeros_(m.bias)
    if isinstance(m, nn.BatchNorm1d):
        nn.init.ones_(m.weight)
        nn.init.zeros_(m.bias)

def get_parameter_number(net):
    total_num = sum(p.numel() for p in net.parameters())
    trainable_num = sum(p.numel() for p in net.parameters() if p.requires_grad)
    return {'Total': total_num, 'Trainable': trainable_num}

#################### For Code Test ##################################
## before running the training codes, verify the network architecture.
# if __name__ == '__main__':
#     mrsnet = MRFNet()
#     mrsnet.apply(weights_init)
#     print(mrsnet.state_dict)
#     print(get_parameter_number(mrsnet))
#     x_r = torch.randn(2,1,1000, dtype=torch.float)
#     x_i = torch.randn(2,1,1000, dtype=torch.float)
#     print('input' + str(x_r.size()))
#     print(x_r.dtype)
#     y = mrsnet(x_r, x_i)
#     print('output'+str(y.size()))