Datasets
Models
Downloads: 1
Diff of
/ecgtoBR/BRnet.py
[000000]
..
[c0487b]
Switch to unified view
| a | b/ecgtoBR/BRnet.py | ||
|---|---|---|---|
| 1 | import torch.nn as nn |
||
| 2 | import torch |
||
| 3 | import torch.nn.functional as F |
||
| 4 | from torch.autograd import Variable |
||
| 5 | |||
| 6 | import numpy as np |
||
| 7 | |||
| 8 | class IncResBlock(nn.Module): ### Inception Resblocks |
||
| 9 | |||
| 10 | def __init__(self, inplanes, planes, convstr=1, convsize = 15, convpadding = 7): |
||
| 11 | |||
| 12 | super(IncResBlock, self).__init__() |
||
| 13 | |||
| 14 | self.Inputconv1x1 = nn.Conv1d(inplanes, planes, kernel_size=1, stride = 1, bias=False) |
||
| 15 | |||
| 16 | self.conv1_1 = nn.Sequential( |
||
| 17 | nn.Conv1d(in_channels = inplanes,out_channels = planes//4,kernel_size = convsize,stride = convstr,padding = convpadding), |
||
| 18 | nn.BatchNorm1d(planes//4)) |
||
| 19 | |||
| 20 | self.conv1_2 = nn.Sequential( |
||
| 21 | nn.Conv1d(inplanes, planes//4, kernel_size=1, stride = convstr, padding=0, bias=False), |
||
| 22 | nn.BatchNorm1d(planes//4), |
||
| 23 | nn.LeakyReLU(0.2,), |
||
| 24 | nn.Conv1d(in_channels = planes//4,out_channels = planes//4,kernel_size = convsize+2,stride = convstr,padding = convpadding+1), |
||
| 25 | nn.BatchNorm1d(planes//4)) |
||
| 26 | |||
| 27 | self.conv1_3 = nn.Sequential( |
||
| 28 | nn.Conv1d(inplanes, planes//4, kernel_size=1, stride = convstr, padding=0, bias=False), |
||
| 29 | nn.BatchNorm1d(planes//4), |
||
| 30 | nn.LeakyReLU(0.2,), |
||
| 31 | nn.Conv1d(in_channels = planes//4,out_channels = planes//4,kernel_size = convsize+4,stride = convstr,padding = convpadding+2), |
||
| 32 | nn.BatchNorm1d(planes//4)) |
||
| 33 | |||
| 34 | self.conv1_4 = nn.Sequential( |
||
| 35 | nn.Conv1d(inplanes, planes//4, kernel_size=1, stride = convstr, padding=0, bias=False), |
||
| 36 | nn.BatchNorm1d(planes//4), |
||
| 37 | nn.LeakyReLU(0.2,), |
||
| 38 | nn.Conv1d(in_channels = planes//4,out_channels = planes//4,kernel_size = convsize+6,stride = convstr,padding = convpadding+3), |
||
| 39 | nn.BatchNorm1d(planes//4)) |
||
| 40 | |||
| 41 | self.relu = nn.ReLU() |
||
| 42 | |||
| 43 | |||
| 44 | def forward(self, x): |
||
| 45 | |||
| 46 | residual = self.Inputconv1x1(x) |
||
| 47 | |||
| 48 | c1 = self.conv1_1(x) |
||
| 49 | c2 = self.conv1_2(x) |
||
| 50 | c3 = self.conv1_3(x) |
||
| 51 | c4 = self.conv1_4(x) |
||
| 52 | |||
| 53 | out = torch.cat([c1,c2,c3,c4],1) |
||
| 54 | out += residual |
||
| 55 | return self.relu(out) |
||
| 56 | |||
| 57 | class IncUNet (nn.Module): ### Inception Unet |
||
| 58 | |||
| 59 | def __init__(self, in_shape): |
||
| 60 | |||
| 61 | super(IncUNet, self).__init__() |
||
| 62 | in_channels, height, width = in_shape |
||
| 63 | |||
| 64 | self.e1 = nn.Sequential( |
||
| 65 | nn.Conv1d(in_channels, 64, kernel_size=4, stride=2,padding=1), |
||
| 66 | nn.BatchNorm1d(64), |
||
| 67 | nn.LeakyReLU(0.2,), |
||
| 68 | IncResBlock(64,64)) |
||
| 69 | |||
| 70 | self.e2 = nn.Sequential( |
||
| 71 | nn.LeakyReLU(0.2,inplace=True), |
||
| 72 | nn.Conv1d(64, 128, kernel_size=4, stride=2,padding=1), |
||
| 73 | nn.BatchNorm1d(128), |
||
| 74 | IncResBlock(128,128)) |
||
| 75 | |||
| 76 | self.e2add = nn.Sequential( |
||
| 77 | nn.Conv1d(128, 128, kernel_size=3, stride=1,padding=1), |
||
| 78 | nn.BatchNorm1d(128)) |
||
| 79 | |||
| 80 | self.e3 = nn.Sequential( |
||
| 81 | nn.LeakyReLU(0.2,inplace=True), |
||
| 82 | nn.Conv1d(128, 128, kernel_size=3, stride=1,padding=1), |
||
| 83 | nn.BatchNorm1d(128), |
||
| 84 | nn.LeakyReLU(0.2,), |
||
| 85 | nn.Conv1d(128,256, kernel_size=4, stride=2,padding=1), |
||
| 86 | nn.BatchNorm1d(256), |
||
| 87 | IncResBlock(256,256)) |
||
| 88 | |||
| 89 | self.e4 = nn.Sequential( |
||
| 90 | nn.LeakyReLU(0.2,), |
||
| 91 | nn.Conv1d(256,256, kernel_size=4 , stride=1 , padding=1), |
||
| 92 | nn.BatchNorm1d(256), |
||
| 93 | nn.LeakyReLU(0.2,inplace=True), |
||
| 94 | nn.Conv1d(256,512, kernel_size=4, stride=2,padding=2), |
||
| 95 | nn.BatchNorm1d(512), |
||
| 96 | IncResBlock(512,512)) |
||
| 97 | |||
| 98 | self.e4add = nn.Sequential( |
||
| 99 | nn.LeakyReLU(0.2,), |
||
| 100 | nn.Conv1d(512,512, kernel_size=3, stride=1,padding=1), |
||
| 101 | nn.BatchNorm1d(512)) |
||
| 102 | |||
| 103 | self.e5 = nn.Sequential( |
||
| 104 | nn.LeakyReLU(0.2,inplace=True), |
||
| 105 | nn.Conv1d(512,512, kernel_size=3, stride=1,padding=1), |
||
| 106 | nn.BatchNorm1d(512), |
||
| 107 | nn.LeakyReLU(0.2,), |
||
| 108 | nn.Conv1d(512,512, kernel_size=4, stride=2,padding=1), |
||
| 109 | nn.BatchNorm1d(512), |
||
| 110 | IncResBlock(512,512)) |
||
| 111 | |||
| 112 | self.e6 = nn.Sequential( |
||
| 113 | nn.LeakyReLU(0.2,), |
||
| 114 | nn.Conv1d(512,512, kernel_size=3, stride=1,padding=1), |
||
| 115 | nn.BatchNorm1d(512), |
||
| 116 | nn.LeakyReLU(0.2,inplace=True), |
||
| 117 | nn.Conv1d(512,512, kernel_size=4, stride=2,padding=1), |
||
| 118 | nn.BatchNorm1d(512), |
||
| 119 | IncResBlock(512,512)) |
||
| 120 | |||
| 121 | self.e6add = nn.Sequential( |
||
| 122 | nn.Conv1d(512,512, kernel_size=3, stride=1,padding=1), |
||
| 123 | nn.BatchNorm1d(512)) |
||
| 124 | |||
| 125 | self.e7 = nn.Sequential( |
||
| 126 | nn.LeakyReLU(0.2,inplace=True), |
||
| 127 | nn.Conv1d(512,512, kernel_size=3, stride=1,padding=1), |
||
| 128 | nn.BatchNorm1d(512), |
||
| 129 | nn.LeakyReLU(0.2,), |
||
| 130 | nn.Conv1d(512,512, kernel_size=4, stride=2,padding=1), |
||
| 131 | nn.BatchNorm1d(512), |
||
| 132 | IncResBlock(512,512)) |
||
| 133 | |||
| 134 | self.e8 = nn.Sequential( |
||
| 135 | nn.LeakyReLU(0.2,), |
||
| 136 | nn.Conv1d(512,512, kernel_size=4, stride=1,padding=1), |
||
| 137 | nn.BatchNorm1d(512), |
||
| 138 | nn.LeakyReLU(0.2,inplace=True), |
||
| 139 | nn.Conv1d(512,512, kernel_size=4, stride=2,padding=1), |
||
| 140 | nn.BatchNorm1d(512)) |
||
| 141 | |||
| 142 | |||
| 143 | self.d1 = nn.Sequential( |
||
| 144 | nn.LeakyReLU(0.2,), |
||
| 145 | nn.ConvTranspose1d(512, 512, kernel_size=4, stride=2,padding=1), |
||
| 146 | nn.BatchNorm1d(512), |
||
| 147 | nn.LeakyReLU(0.2,), |
||
| 148 | nn.ConvTranspose1d(512, 512, kernel_size=4, stride=1,padding =1), |
||
| 149 | nn.BatchNorm1d(512), |
||
| 150 | IncResBlock(512,512)) |
||
| 151 | |||
| 152 | self.d2 = nn.Sequential( |
||
| 153 | nn.LeakyReLU(0.2,), |
||
| 154 | nn.ConvTranspose1d(1024, 512, kernel_size=4, stride=2,padding=1), |
||
| 155 | nn.BatchNorm1d(512), |
||
| 156 | nn.LeakyReLU(0.2,), |
||
| 157 | nn.ConvTranspose1d(512, 512, kernel_size=3, stride=1,padding=1), |
||
| 158 | nn.BatchNorm1d(512), |
||
| 159 | IncResBlock(512,512)) |
||
| 160 | |||
| 161 | self.d3 = nn.Sequential( |
||
| 162 | nn.LeakyReLU(0.2,), |
||
| 163 | nn.ConvTranspose1d(1024, 512, kernel_size=3, stride=1,padding=1), |
||
| 164 | nn.BatchNorm1d(512), |
||
| 165 | nn.Dropout(p=0.5), |
||
| 166 | IncResBlock(512,512)) |
||
| 167 | |||
| 168 | self.d4 = nn.Sequential( |
||
| 169 | nn.LeakyReLU(0.2,), |
||
| 170 | nn.ConvTranspose1d(1024, 512, kernel_size=4, stride=2,padding=1), |
||
| 171 | nn.BatchNorm1d(512), |
||
| 172 | nn.LeakyReLU(0.2,), |
||
| 173 | nn.ConvTranspose1d(512, 512, kernel_size=3, stride=1,padding=1), |
||
| 174 | nn.BatchNorm1d(512), |
||
| 175 | IncResBlock(512,512)) |
||
| 176 | |||
| 177 | self.d5 = nn.Sequential( |
||
| 178 | nn.LeakyReLU(0.2,), |
||
| 179 | nn.ConvTranspose1d(1024, 512, kernel_size=4, stride=2,padding=1), |
||
| 180 | nn.BatchNorm1d(512), |
||
| 181 | nn.LeakyReLU(0.2,), |
||
| 182 | nn.ConvTranspose1d(512, 512, kernel_size=3, stride=1,padding=1), |
||
| 183 | nn.BatchNorm1d(512), |
||
| 184 | IncResBlock(512,512)) |
||
| 185 | |||
| 186 | self.d6 = nn.Sequential( |
||
| 187 | nn.LeakyReLU(0.2,), |
||
| 188 | nn.ConvTranspose1d(1024, 512, kernel_size=3, stride=1,padding=1), |
||
| 189 | nn.BatchNorm1d(512), |
||
| 190 | IncResBlock(512,512)) |
||
| 191 | |||
| 192 | self.d7 = nn.Sequential( |
||
| 193 | nn.LeakyReLU(0.2,), |
||
| 194 | nn.ConvTranspose1d(1024, 256, kernel_size=4, stride=2,padding=1), |
||
| 195 | nn.BatchNorm1d(256), |
||
| 196 | nn.LeakyReLU(0.2,), |
||
| 197 | nn.ConvTranspose1d(256, 256, kernel_size=3, stride=1,padding=1), |
||
| 198 | nn.BatchNorm1d(256), |
||
| 199 | IncResBlock(256,256)) |
||
| 200 | |||
| 201 | self.d8 = nn.Sequential( |
||
| 202 | nn.LeakyReLU(0.2,), |
||
| 203 | nn.ConvTranspose1d(512, 128, kernel_size=4, stride=2,padding=1), |
||
| 204 | nn.BatchNorm1d(128), |
||
| 205 | nn.LeakyReLU(0.2,), |
||
| 206 | nn.ConvTranspose1d(128, 128, kernel_size=3, stride=1,padding=1), |
||
| 207 | nn.BatchNorm1d(128)) |
||
| 208 | |||
| 209 | self.d9 = nn.Sequential( |
||
| 210 | nn.LeakyReLU(0.2,), |
||
| 211 | nn.ConvTranspose1d(256, 128, kernel_size=3, stride=1,padding=1), |
||
| 212 | nn.BatchNorm1d(128)) |
||
| 213 | |||
| 214 | self.d10 = nn.Sequential( |
||
| 215 | nn.LeakyReLU(0.2,), |
||
| 216 | nn.ConvTranspose1d(256, 64, kernel_size=3, stride=1,padding=1), |
||
| 217 | nn.BatchNorm1d(64)) |
||
| 218 | |||
| 219 | self.out_l = nn.Sequential( |
||
| 220 | nn.LeakyReLU(0.2,), |
||
| 221 | nn.ConvTranspose1d(256, in_channels, kernel_size=3, stride=1,padding=1)) |
||
| 222 | |||
| 223 | |||
| 224 | def forward(self, x): |
||
| 225 | |||
| 226 | ### Encoder |
||
| 227 | en1 = self.e1(x) |
||
| 228 | en2 = self.e2(en1) |
||
| 229 | en2add = self.e2add(en2) |
||
| 230 | en3 = self.e3(en2add) |
||
| 231 | en4 = self.e4(en3) |
||
| 232 | en4add = self.e4add(en4) |
||
| 233 | en5 = self.e5(en4add) |
||
| 234 | en6 = self.e6(en5) |
||
| 235 | en6add = self.e6add(en6) |
||
| 236 | en7 = self.e7(en6add) |
||
| 237 | en8 = self.e8(en7) |
||
| 238 | |||
| 239 | |||
| 240 | ### Decoder |
||
| 241 | de1_ = self.d1(en8) |
||
| 242 | de1 = torch.cat([en7,de1_],1) |
||
| 243 | de2_ = self.d2(de1) |
||
| 244 | de2 = torch.cat([en6add,de2_],1) |
||
| 245 | de3_ = self.d3(de2) |
||
| 246 | de3 = torch.cat([en6,de3_],1) |
||
| 247 | de4_ = self.d4(de3) |
||
| 248 | de4 = torch.cat([en5,de4_],1) |
||
| 249 | de5_ = self.d5(de4) |
||
| 250 | de5_ = nn.ConstantPad1d((0,1),0)(de5_) |
||
| 251 | de5 = torch.cat([en4add,de5_],1) |
||
| 252 | de6_ = self.d6(de5) |
||
| 253 | de6 = torch.cat([en4,de6_],1) |
||
| 254 | de7_ = self.d7(de6) |
||
| 255 | de7_ = de7_[:,:,:-1] |
||
| 256 | de7 = torch.cat([en3,de7_],1) |
||
| 257 | de8 = self.d8(de7) |
||
| 258 | de8_ = self.d8(de7) |
||
| 259 | de8 = torch.cat([en2add,de8_],1) |
||
| 260 | |||
| 261 | return self.out_l(de8) |
||
| 262 | |||
| 263 |