PreOpNet/models.py at main · douyang/PreOpNet · 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
from torch import nn
import torch
import torch.nn.functional as F

# +
from torch import nn
from collections import OrderedDict


class Bottleneck(nn.Module):
    def __init__(self,in_channel,out_channel,expansion,activation,stride=1,padding = 1):
        super(Bottleneck, self).__init__()
        self.stride=stride
        self.conv1 = nn.Conv1d(in_channel,in_channel*expansion,kernel_size = 1)
        self.conv2 = nn.Conv1d(in_channel*expansion,in_channel*expansion,kernel_size = 3, groups = in_channel*expansion,
                               padding=padding,stride = stride)
        self.conv3 = nn.Conv1d(in_channel*expansion,out_channel,kernel_size = 1, stride =1)
        self.b0 = nn.BatchNorm1d(in_channel*expansion)
        self.b1 =  nn.BatchNorm1d(in_channel*expansion)
        self.d = nn.Dropout()
        self.act = activation()
    def forward(self,x):
        if self.stride == 1:
            y = self.act(self.b0(self.conv1(x)))
            y = self.act(self.b1(self.conv2(y)))
            y = self.conv3(y)
            y = self.d(y)
            y = x+y
            return y
        else:
            y = self.act(self.b0(self.conv1(x)))
            y = self.act(self.b1(self.conv2(y)))
            y = self.conv3(y)
            return y

from torch import nn
from collections import OrderedDict

class MBConv(nn.Module):
    def __init__(self,in_channel,out_channels,expansion,layers,activation=nn.ReLU6,stride = 2):
        super(MBConv, self).__init__()
        self.stack = OrderedDict()
        for i in range(0,layers-1):
            self.stack['s'+str(i)] = Bottleneck(in_channel,in_channel,expansion,activation)
            #self.stack['a'+str(i)] = activation()
        self.stack['s'+str(layers+1)] = Bottleneck(in_channel,out_channels,expansion,activation,stride=stride)
        # self.stack['a'+str(layers+1)] = activation()
        self.stack = nn.Sequential(self.stack)

        self.bn = nn.BatchNorm1d(out_channels)
    def forward(self,x):
        x = self.stack(x)
        return self.bn(x)


"""def MBConv(in_channel,out_channels,expansion,layers,activation=nn.ReLU6,stride = 2):
    stack = OrderedDict()
    for i in range(0,layers-1):
        stack['b'+str(i)] = Bottleneck(in_channel,in_channel,expansion,activation)
    stack['b'+str(layers)] = Bottleneck(in_channel,out_channels,expansion,activation,stride=stride)
    return nn.Sequential(stack)"""


class EffNet(nn.Module):

    def __init__(self,num_additional_features = 0,depth = [1,2,2,3,3,3,3],channels = [32,16,24,40,80,112,192,320,1280],
                dilation = 1,stride = 2,expansion = 6):
        super(EffNet, self).__init__()
        print("depth ",depth)
        self.stage1 = nn.Conv1d(12, channels[0], kernel_size=3, stride=stride, padding=1,dilation = dilation) #1 conv
        self.b0 = nn.BatchNorm1d(channels[0])
        self.stage2 = MBConv(channels[0], channels[1], expansion, depth[0], stride=2)# 16 #input, output, depth # 3 conv
        self.stage3 = MBConv(channels[1], channels[2], expansion, depth[1], stride=2)# 24 # 4 conv # d 2
        self.Pool = nn.MaxPool1d(3, stride=1, padding=1) #
        self.stage4 = MBConv(channels[2], channels[3], expansion, depth[2], stride=2)# 40 # 4 conv # d 2
        self.stage5 = MBConv(channels[3], channels[4], expansion, depth[3], stride=2)# 80 # 5 conv # d
        self.stage6 = MBConv(channels[4], channels[5], expansion, depth[4], stride=2)# 112 # 5 conv
        self.stage7 = MBConv(channels[5], channels[6], expansion, depth[5], stride=2)# 192 # 5 conv
        self.stage8 = MBConv(channels[6], channels[7], expansion, depth[6], stride=2)# 320 # 5 conv

        self.stage9 = nn.Conv1d(channels[7], channels[8], kernel_size=1)
        self.AAP = nn.AdaptiveAvgPool1d(1)
        self.act = nn.ReLU()
        self.drop = nn.Dropout()
        self.num_additional_features = num_additional_features
        self.fc = nn.Linear(channels[8] + num_additional_features, 1)


    def forward(self, x):
        if self.num_additional_features >0:
            x,additional = x
        # N x 12 x 2500
        x = self.b0(self.stage1(x))
        # N x 32 x 1250
        x = self.stage2(x)
        # N x 16 x 625
        x = self.stage3(x)
        # N x 24 x 313
        x = self.Pool(x)
        # N x 24 x 313

        x = self.stage4(x)
        # N x 40 x 157
        x = self.stage5(x)
        # N x 80 x 79
        x = self.stage6(x)
        # N x 112 x 40
        x = self.Pool(x)
        # N x 192 x 20

        x = self.stage7(x)
        # N x 320 x 10
        x = self.stage8(x)
        x = self.stage9(x)
        # N x 1280 x 10
        x = self.act(self.AAP(x)[:,:,0])
        # N x 1280
        x = self.drop(x)
        if self.num_additional_features >0:
            x = torch.cat((x,additional),1)
        x = self.fc(x)
        # N x 1
        return x