Skip to content

odak.learn.models

odak.learn.models

Provides necessary definitions for components used in machine learning and deep learning.

convolution_layer

Bases: Module

A convolution layer.

Source code in odak/learn/models/components.py 
 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
class convolution_layer(torch.nn.Module):
    """
    A convolution layer.
    """
    def __init__(
                 self,
                 input_channels = 2,
                 output_channels = 2,
                 kernel_size = 3,
                 bias = False,
                 activation = torch.nn.ReLU()
                ):
        """
        A convolutional layer class.


        Parameters
        ----------
        input_channels  : int
                          Number of input channels.
        output_channels : int
                          Number of output channels.
        kernel_size     : int
                          Kernel size.
        bias            : bool 
                          Set to True to let convolutional layers have bias term.
        activation      : torch.nn
                          Nonlinear activation layer to be used. If None, uses torch.nn.ReLU().
        """
        super().__init__()
        self.activation = activation
        self.model = torch.nn.Sequential(
                                         torch.nn.Conv2d(
                                                         input_channels,
                                                         output_channels,
                                                         kernel_size = kernel_size,
                                                         padding = kernel_size // 2,
                                                         bias = bias
                                                        ),
                                         torch.nn.BatchNorm2d(output_channels),
                                         self.activation
                                        )


    def forward(self, x):
        """
        Forward model.

        Parameters
        ----------
        x             : torch.tensor
                        Input data.


        Returns
        ----------
        result        : torch.tensor
                        Estimated output.      
        """
        result = self.model(x)
        return result

__init__(input_channels=2, output_channels=2, kernel_size=3, bias=False, activation=torch.nn.ReLU())

A convolutional layer class.

Parameters:

  • input_channels – 
              Number of input channels.
  • output_channels (int, default: 2 ) – 
              Number of output channels.
  • kernel_size – 
              Kernel size.
  • bias – 
              Set to True to let convolutional layers have bias term.
  • activation – 
              Nonlinear activation layer to be used. If None, uses torch.nn.ReLU().
Source code in odak/learn/models/components.py 
 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
def __init__(
             self,
             input_channels = 2,
             output_channels = 2,
             kernel_size = 3,
             bias = False,
             activation = torch.nn.ReLU()
            ):
    """
    A convolutional layer class.


    Parameters
    ----------
    input_channels  : int
                      Number of input channels.
    output_channels : int
                      Number of output channels.
    kernel_size     : int
                      Kernel size.
    bias            : bool 
                      Set to True to let convolutional layers have bias term.
    activation      : torch.nn
                      Nonlinear activation layer to be used. If None, uses torch.nn.ReLU().
    """
    super().__init__()
    self.activation = activation
    self.model = torch.nn.Sequential(
                                     torch.nn.Conv2d(
                                                     input_channels,
                                                     output_channels,
                                                     kernel_size = kernel_size,
                                                     padding = kernel_size // 2,
                                                     bias = bias
                                                    ),
                                     torch.nn.BatchNorm2d(output_channels),
                                     self.activation
                                    )

forward(x)

Forward model.

Parameters:

  • x – 
            Input data.

Returns:

  • result ( tensor ) –

    Estimated output.

Source code in odak/learn/models/components.py 
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
def forward(self, x):
    """
    Forward model.

    Parameters
    ----------
    x             : torch.tensor
                    Input data.


    Returns
    ----------
    result        : torch.tensor
                    Estimated output.      
    """
    result = self.model(x)
    return result

double_convolution

Bases: Module

A double convolution layer.

Source code in odak/learn/models/components.py 
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
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
class double_convolution(torch.nn.Module):
    """
    A double convolution layer.
    """
    def __init__(
                 self,
                 input_channels = 2,
                 mid_channels = None,
                 output_channels = 2,
                 kernel_size = 3, 
                 bias = False,
                 activation = torch.nn.ReLU()
                ):
        """
        Double convolution model.


        Parameters
        ----------
        input_channels  : int
                          Number of input channels.
        mid_channels    : int
                          Number of channels in the hidden layer between two convolutions.
        output_channels : int
                          Number of output channels.
        kernel_size     : int
                          Kernel size.
        bias            : bool 
                          Set to True to let convolutional layers have bias term.
        activation      : torch.nn
                          Nonlinear activation layer to be used. If None, uses torch.nn.ReLU().
        """
        super().__init__()
        if isinstance(mid_channels, type(None)):
            mid_channels = output_channels
        self.activation = activation
        self.model = torch.nn.Sequential(
                                         convolution_layer(
                                                           input_channels = input_channels,
                                                           output_channels = mid_channels,
                                                           kernel_size = kernel_size,
                                                           bias = bias,
                                                           activation = self.activation
                                                          ),
                                         convolution_layer(
                                                           input_channels = mid_channels,
                                                           output_channels = output_channels,
                                                           kernel_size = kernel_size,
                                                           bias = bias,
                                                           activation = self.activation
                                                          )
                                        )


    def forward(self, x):
        """
        Forward model.

        Parameters
        ----------
        x             : torch.tensor
                        Input data.


        Returns
        ----------
        result        : torch.tensor
                        Estimated output.      
        """
        result = self.model(x)
        return result

__init__(input_channels=2, mid_channels=None, output_channels=2, kernel_size=3, bias=False, activation=torch.nn.ReLU())

Double convolution model.

Parameters:

  • input_channels – 
              Number of input channels.
  • mid_channels – 
              Number of channels in the hidden layer between two convolutions.
  • output_channels (int, default: 2 ) – 
              Number of output channels.
  • kernel_size – 
              Kernel size.
  • bias – 
              Set to True to let convolutional layers have bias term.
  • activation – 
              Nonlinear activation layer to be used. If None, uses torch.nn.ReLU().
Source code in odak/learn/models/components.py 
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
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
def __init__(
             self,
             input_channels = 2,
             mid_channels = None,
             output_channels = 2,
             kernel_size = 3, 
             bias = False,
             activation = torch.nn.ReLU()
            ):
    """
    Double convolution model.


    Parameters
    ----------
    input_channels  : int
                      Number of input channels.
    mid_channels    : int
                      Number of channels in the hidden layer between two convolutions.
    output_channels : int
                      Number of output channels.
    kernel_size     : int
                      Kernel size.
    bias            : bool 
                      Set to True to let convolutional layers have bias term.
    activation      : torch.nn
                      Nonlinear activation layer to be used. If None, uses torch.nn.ReLU().
    """
    super().__init__()
    if isinstance(mid_channels, type(None)):
        mid_channels = output_channels
    self.activation = activation
    self.model = torch.nn.Sequential(
                                     convolution_layer(
                                                       input_channels = input_channels,
                                                       output_channels = mid_channels,
                                                       kernel_size = kernel_size,
                                                       bias = bias,
                                                       activation = self.activation
                                                      ),
                                     convolution_layer(
                                                       input_channels = mid_channels,
                                                       output_channels = output_channels,
                                                       kernel_size = kernel_size,
                                                       bias = bias,
                                                       activation = self.activation
                                                      )
                                    )

forward(x)

Forward model.

Parameters:

  • x – 
            Input data.

Returns:

  • result ( tensor ) –

    Estimated output.

Source code in odak/learn/models/components.py 
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
def forward(self, x):
    """
    Forward model.

    Parameters
    ----------
    x             : torch.tensor
                    Input data.


    Returns
    ----------
    result        : torch.tensor
                    Estimated output.      
    """
    result = self.model(x)
    return result

downsample_layer

Bases: Module

A downscaling component followed by a double convolution.

Source code in odak/learn/models/components.py 
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
class downsample_layer(torch.nn.Module):
    """
    A downscaling component followed by a double convolution.
    """
    def __init__(
                 self,
                 input_channels,
                 output_channels,
                 kernel_size = 3,
                 bias = False,
                 activation = torch.nn.ReLU()
                ):
        """
        A downscaling component with a double convolution.

        Parameters
        ----------
        input_channels  : int
                          Number of input channels.
        output_channels : int
                          Number of output channels.
        kernel_size     : int
                          Kernel size.
        bias            : bool 
                          Set to True to let convolutional layers have bias term.
        activation      : torch.nn
                          Nonlinear activation layer to be used. If None, uses torch.nn.ReLU().
        """
        super().__init__()
        self.maxpool_conv = torch.nn.Sequential(
                                                torch.nn.MaxPool2d(2),
                                                double_convolution(
                                                                   input_channels = input_channels,
                                                                   mid_channels = output_channels,
                                                                   output_channels = output_channels,
                                                                   kernel_size = kernel_size,
                                                                   bias = bias,
                                                                   activation = activation
                                                                  )
                                               )

    def forward(self, x):
        """
        Forward model.

        Parameters
        ----------
        x              : torch.tensor
                         First input data.



        Returns
        ----------
        result        : torch.tensor
                        Estimated output.      
        """
        result = self.maxpool_conv(x)
        return result

__init__(input_channels, output_channels, kernel_size=3, bias=False, activation=torch.nn.ReLU())

A downscaling component with a double convolution.

Parameters:

  • input_channels – 
              Number of input channels.
  • output_channels (int) – 
              Number of output channels.
  • kernel_size – 
              Kernel size.
  • bias – 
              Set to True to let convolutional layers have bias term.
  • activation – 
              Nonlinear activation layer to be used. If None, uses torch.nn.ReLU().
Source code in odak/learn/models/components.py 
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
def __init__(
             self,
             input_channels,
             output_channels,
             kernel_size = 3,
             bias = False,
             activation = torch.nn.ReLU()
            ):
    """
    A downscaling component with a double convolution.

    Parameters
    ----------
    input_channels  : int
                      Number of input channels.
    output_channels : int
                      Number of output channels.
    kernel_size     : int
                      Kernel size.
    bias            : bool 
                      Set to True to let convolutional layers have bias term.
    activation      : torch.nn
                      Nonlinear activation layer to be used. If None, uses torch.nn.ReLU().
    """
    super().__init__()
    self.maxpool_conv = torch.nn.Sequential(
                                            torch.nn.MaxPool2d(2),
                                            double_convolution(
                                                               input_channels = input_channels,
                                                               mid_channels = output_channels,
                                                               output_channels = output_channels,
                                                               kernel_size = kernel_size,
                                                               bias = bias,
                                                               activation = activation
                                                              )
                                           )

forward(x)

Forward model.

Parameters:

  • x – 
             First input data.

Returns:

  • result ( tensor ) –

    Estimated output.

Source code in odak/learn/models/components.py 
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
def forward(self, x):
    """
    Forward model.

    Parameters
    ----------
    x              : torch.tensor
                     First input data.



    Returns
    ----------
    result        : torch.tensor
                    Estimated output.      
    """
    result = self.maxpool_conv(x)
    return result

non_local_layer

Bases: Module

Self-Attention Layer [zi = Wzyi + xi] (non-local block : ref https://arxiv.org/abs/1711.07971)

Source code in odak/learn/models/components.py 
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
class non_local_layer(torch.nn.Module):
    """
    Self-Attention Layer [zi = Wzyi + xi] (non-local block : ref https://arxiv.org/abs/1711.07971)
    """
    def __init__(
                 self,
                 input_channels = 1024,
                 bottleneck_channels = 512,
                 kernel_size = 1,
                 bias = False,
                ):
        """

        Parameters
        ----------
        input_channels      : int
                              Number of input channels.
        bottleneck_channels : int
                              Number of middle channels.
        kernel_size         : int
                              Kernel size.
        bias                : bool 
                              Set to True to let convolutional layers have bias term.
        """
        super(non_local_layer, self).__init__()
        self.input_channels = input_channels
        self.bottleneck_channels = bottleneck_channels
        self.g = torch.nn.Conv2d(
                                 self.input_channels, 
                                 self.bottleneck_channels,
                                 kernel_size = kernel_size,
                                 padding = kernel_size // 2,
                                 bias = bias
                                )
        self.W_z = torch.nn.Sequential(
                                       torch.nn.Conv2d(
                                                       self.bottleneck_channels,
                                                       self.input_channels, 
                                                       kernel_size = kernel_size,
                                                       bias = bias,
                                                       padding = kernel_size // 2
                                                      ),
                                       torch.nn.BatchNorm2d(self.input_channels)
                                      )
        torch.nn.init.constant_(self.W_z[1].weight, 0)   
        torch.nn.init.constant_(self.W_z[1].bias, 0)


    def forward(self, x):
        """
        Forward model [zi = Wzyi + xi]

        Parameters
        ----------
        x               : torch.tensor
                          First input data.                       


        Returns
        ----------
        z               : torch.tensor
                          Estimated output.
        """
        batch_size, channels, height, width = x.size()
        theta = x.view(batch_size, channels, -1).permute(0, 2, 1)
        phi = x.view(batch_size, channels, -1).permute(0, 2, 1)
        g = self.g(x).view(batch_size, self.bottleneck_channels, -1).permute(0, 2, 1)
        attn = torch.bmm(theta, phi.transpose(1, 2)) / (height * width)
        attn = torch.nn.functional.softmax(attn, dim=-1)
        y = torch.bmm(attn, g).permute(0, 2, 1).contiguous().view(batch_size, self.bottleneck_channels, height, width)
        W_y = self.W_z(y)
        z = W_y + x
        return z

__init__(input_channels=1024, bottleneck_channels=512, kernel_size=1, bias=False)

Parameters:

  • input_channels – 
                  Number of input channels.
  • bottleneck_channels (int, default: 512 ) – 
                  Number of middle channels.
  • kernel_size – 
                  Kernel size.
  • bias – 
                  Set to True to let convolutional layers have bias term.
Source code in odak/learn/models/components.py 
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
def __init__(
             self,
             input_channels = 1024,
             bottleneck_channels = 512,
             kernel_size = 1,
             bias = False,
            ):
    """

    Parameters
    ----------
    input_channels      : int
                          Number of input channels.
    bottleneck_channels : int
                          Number of middle channels.
    kernel_size         : int
                          Kernel size.
    bias                : bool 
                          Set to True to let convolutional layers have bias term.
    """
    super(non_local_layer, self).__init__()
    self.input_channels = input_channels
    self.bottleneck_channels = bottleneck_channels
    self.g = torch.nn.Conv2d(
                             self.input_channels, 
                             self.bottleneck_channels,
                             kernel_size = kernel_size,
                             padding = kernel_size // 2,
                             bias = bias
                            )
    self.W_z = torch.nn.Sequential(
                                   torch.nn.Conv2d(
                                                   self.bottleneck_channels,
                                                   self.input_channels, 
                                                   kernel_size = kernel_size,
                                                   bias = bias,
                                                   padding = kernel_size // 2
                                                  ),
                                   torch.nn.BatchNorm2d(self.input_channels)
                                  )
    torch.nn.init.constant_(self.W_z[1].weight, 0)   
    torch.nn.init.constant_(self.W_z[1].bias, 0)

forward(x)

Forward model [zi = Wzyi + xi]

Parameters:

  • x – 
              First input data.

Returns:

  • z ( tensor ) –

    Estimated output.

Source code in odak/learn/models/components.py 
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
def forward(self, x):
    """
    Forward model [zi = Wzyi + xi]

    Parameters
    ----------
    x               : torch.tensor
                      First input data.                       


    Returns
    ----------
    z               : torch.tensor
                      Estimated output.
    """
    batch_size, channels, height, width = x.size()
    theta = x.view(batch_size, channels, -1).permute(0, 2, 1)
    phi = x.view(batch_size, channels, -1).permute(0, 2, 1)
    g = self.g(x).view(batch_size, self.bottleneck_channels, -1).permute(0, 2, 1)
    attn = torch.bmm(theta, phi.transpose(1, 2)) / (height * width)
    attn = torch.nn.functional.softmax(attn, dim=-1)
    y = torch.bmm(attn, g).permute(0, 2, 1).contiguous().view(batch_size, self.bottleneck_channels, height, width)
    W_y = self.W_z(y)
    z = W_y + x
    return z

normalization

Bases: Module

A normalization layer.

Source code in odak/learn/models/components.py 
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
class normalization(torch.nn.Module):
    """
    A normalization layer.
    """
    def __init__(
                 self,
                 dim = 1,
                ):
        """
        Normalization layer.


        Parameters
        ----------
        dim             : int
                          Dimension (axis) to normalize.
        """
        super().__init__()
        self.k = torch.nn.Parameter(torch.ones(1, dim, 1, 1))


    def forward(self, x):
        """
        Forward model.

        Parameters
        ----------
        x             : torch.tensor
                        Input data.


        Returns
        ----------
        result        : torch.tensor
                        Estimated output.      
        """
        eps = 1e-5 if x.dtype == torch.float32 else 1e-3
        var = torch.var(x, dim = 1, unbiased = False, keepdim = True)
        mean = torch.mean(x, dim = 1, keepdim = True)
        result =  (x - mean) * (var + eps).rsqrt() * self.k
        return result

__init__(dim=1)

Normalization layer.

Parameters:

  • dim – 
              Dimension (axis) to normalize.
Source code in odak/learn/models/components.py 
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
def __init__(
             self,
             dim = 1,
            ):
    """
    Normalization layer.


    Parameters
    ----------
    dim             : int
                      Dimension (axis) to normalize.
    """
    super().__init__()
    self.k = torch.nn.Parameter(torch.ones(1, dim, 1, 1))

forward(x)

Forward model.

Parameters:

  • x – 
            Input data.

Returns:

  • result ( tensor ) –

    Estimated output.

Source code in odak/learn/models/components.py 
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
def forward(self, x):
    """
    Forward model.

    Parameters
    ----------
    x             : torch.tensor
                    Input data.


    Returns
    ----------
    result        : torch.tensor
                    Estimated output.      
    """
    eps = 1e-5 if x.dtype == torch.float32 else 1e-3
    var = torch.var(x, dim = 1, unbiased = False, keepdim = True)
    mean = torch.mean(x, dim = 1, keepdim = True)
    result =  (x - mean) * (var + eps).rsqrt() * self.k
    return result

residual_attention_layer

Bases: Module

A residual block with an attention layer.

Source code in odak/learn/models/components.py 
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
class residual_attention_layer(torch.nn.Module):
    """
    A residual block with an attention layer.
    """
    def __init__(
                 self,
                 input_channels = 2,
                 output_channels = 2,
                 kernel_size = 1,
                 bias = False,
                 activation = torch.nn.ReLU()
                ):
        """
        An attention layer class.


        Parameters
        ----------
        input_channels  : int or optioal
                          Number of input channels.
        output_channels : int or optional
                          Number of middle channels.
        kernel_size     : int or optional
                          Kernel size.
        bias            : bool or optional
                          Set to True to let convolutional layers have bias term.
        activation      : torch.nn or optional
                          Nonlinear activation layer to be used. If None, uses torch.nn.ReLU().
        """
        super().__init__()
        self.activation = activation
        self.convolution0 = torch.nn.Sequential(
                                                torch.nn.Conv2d(
                                                                input_channels,
                                                                output_channels,
                                                                kernel_size = kernel_size,
                                                                padding = kernel_size // 2,
                                                                bias = bias
                                                               ),
                                                torch.nn.BatchNorm2d(output_channels)
                                               )
        self.convolution1 = torch.nn.Sequential(
                                                torch.nn.Conv2d(
                                                                input_channels,
                                                                output_channels,
                                                                kernel_size = kernel_size,
                                                                padding = kernel_size // 2,
                                                                bias = bias
                                                               ),
                                                torch.nn.BatchNorm2d(output_channels)
                                               )
        self.final_layer = torch.nn.Sequential(
                                               self.activation,
                                               torch.nn.Conv2d(
                                                               output_channels,
                                                               output_channels,
                                                               kernel_size = kernel_size,
                                                               padding = kernel_size // 2,
                                                               bias = bias
                                                              )
                                              )


    def forward(self, x0, x1):
        """
        Forward model.

        Parameters
        ----------
        x0             : torch.tensor
                         First input data.

        x1             : torch.tensor
                         Seconnd input data.


        Returns
        ----------
        result        : torch.tensor
                        Estimated output.      
        """
        y0 = self.convolution0(x0)
        y1 = self.convolution1(x1)
        y2 = torch.add(y0, y1)
        result = self.final_layer(y2) * x0
        return result

__init__(input_channels=2, output_channels=2, kernel_size=1, bias=False, activation=torch.nn.ReLU())

An attention layer class.

Parameters:

  • input_channels – 
              Number of input channels.
  • output_channels (int or optional, default: 2 ) – 
              Number of middle channels.
  • kernel_size – 
              Kernel size.
  • bias – 
              Set to True to let convolutional layers have bias term.
  • activation – 
              Nonlinear activation layer to be used. If None, uses torch.nn.ReLU().
Source code in odak/learn/models/components.py 
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
def __init__(
             self,
             input_channels = 2,
             output_channels = 2,
             kernel_size = 1,
             bias = False,
             activation = torch.nn.ReLU()
            ):
    """
    An attention layer class.


    Parameters
    ----------
    input_channels  : int or optioal
                      Number of input channels.
    output_channels : int or optional
                      Number of middle channels.
    kernel_size     : int or optional
                      Kernel size.
    bias            : bool or optional
                      Set to True to let convolutional layers have bias term.
    activation      : torch.nn or optional
                      Nonlinear activation layer to be used. If None, uses torch.nn.ReLU().
    """
    super().__init__()
    self.activation = activation
    self.convolution0 = torch.nn.Sequential(
                                            torch.nn.Conv2d(
                                                            input_channels,
                                                            output_channels,
                                                            kernel_size = kernel_size,
                                                            padding = kernel_size // 2,
                                                            bias = bias
                                                           ),
                                            torch.nn.BatchNorm2d(output_channels)
                                           )
    self.convolution1 = torch.nn.Sequential(
                                            torch.nn.Conv2d(
                                                            input_channels,
                                                            output_channels,
                                                            kernel_size = kernel_size,
                                                            padding = kernel_size // 2,
                                                            bias = bias
                                                           ),
                                            torch.nn.BatchNorm2d(output_channels)
                                           )
    self.final_layer = torch.nn.Sequential(
                                           self.activation,
                                           torch.nn.Conv2d(
                                                           output_channels,
                                                           output_channels,
                                                           kernel_size = kernel_size,
                                                           padding = kernel_size // 2,
                                                           bias = bias
                                                          )
                                          )

forward(x0, x1)

Forward model.

Parameters:

  • x0 – 
             First input data.
  • x1 – 
             Seconnd input data.

Returns:

  • result ( tensor ) –

    Estimated output.

Source code in odak/learn/models/components.py 
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
def forward(self, x0, x1):
    """
    Forward model.

    Parameters
    ----------
    x0             : torch.tensor
                     First input data.

    x1             : torch.tensor
                     Seconnd input data.


    Returns
    ----------
    result        : torch.tensor
                    Estimated output.      
    """
    y0 = self.convolution0(x0)
    y1 = self.convolution1(x1)
    y2 = torch.add(y0, y1)
    result = self.final_layer(y2) * x0
    return result

residual_layer

Bases: Module

A residual layer.

Source code in odak/learn/models/components.py 
 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
class residual_layer(torch.nn.Module):
    """
    A residual layer.
    """
    def __init__(
                 self,
                 input_channels = 2,
                 mid_channels = 16,
                 kernel_size = 3,
                 bias = False,
                 activation = torch.nn.ReLU()
                ):
        """
        A convolutional layer class.


        Parameters
        ----------
        input_channels  : int
                          Number of input channels.
        mid_channels    : int
                          Number of middle channels.
        kernel_size     : int
                          Kernel size.
        bias            : bool 
                          Set to True to let convolutional layers have bias term.
        activation      : torch.nn
                          Nonlinear activation layer to be used. If None, uses torch.nn.ReLU().
        """
        super().__init__()
        self.activation = activation
        self.convolution = double_convolution(
                                              input_channels,
                                              mid_channels = mid_channels,
                                              output_channels = input_channels,
                                              kernel_size = kernel_size,
                                              bias = bias,
                                              activation = activation
                                             )


    def forward(self, x):
        """
        Forward model.

        Parameters
        ----------
        x             : torch.tensor
                        Input data.


        Returns
        ----------
        result        : torch.tensor
                        Estimated output.      
        """
        x0 = self.convolution(x)
        return x + x0

__init__(input_channels=2, mid_channels=16, kernel_size=3, bias=False, activation=torch.nn.ReLU())

A convolutional layer class.

Parameters:

  • input_channels – 
              Number of input channels.
  • mid_channels – 
              Number of middle channels.
  • kernel_size – 
              Kernel size.
  • bias – 
              Set to True to let convolutional layers have bias term.
  • activation – 
              Nonlinear activation layer to be used. If None, uses torch.nn.ReLU().
Source code in odak/learn/models/components.py 
 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
def __init__(
             self,
             input_channels = 2,
             mid_channels = 16,
             kernel_size = 3,
             bias = False,
             activation = torch.nn.ReLU()
            ):
    """
    A convolutional layer class.


    Parameters
    ----------
    input_channels  : int
                      Number of input channels.
    mid_channels    : int
                      Number of middle channels.
    kernel_size     : int
                      Kernel size.
    bias            : bool 
                      Set to True to let convolutional layers have bias term.
    activation      : torch.nn
                      Nonlinear activation layer to be used. If None, uses torch.nn.ReLU().
    """
    super().__init__()
    self.activation = activation
    self.convolution = double_convolution(
                                          input_channels,
                                          mid_channels = mid_channels,
                                          output_channels = input_channels,
                                          kernel_size = kernel_size,
                                          bias = bias,
                                          activation = activation
                                         )

forward(x)

Forward model.

Parameters:

  • x – 
            Input data.

Returns:

  • result ( tensor ) –

    Estimated output.

Source code in odak/learn/models/components.py 
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
def forward(self, x):
    """
    Forward model.

    Parameters
    ----------
    x             : torch.tensor
                    Input data.


    Returns
    ----------
    result        : torch.tensor
                    Estimated output.      
    """
    x0 = self.convolution(x)
    return x + x0

upsample_layer

Bases: Module

An upsampling convolutional layer.

Source code in odak/learn/models/components.py 
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
class upsample_layer(torch.nn.Module):
    """
    An upsampling convolutional layer.
    """
    def __init__(
                 self,
                 input_channels,
                 output_channels,
                 kernel_size = 3,
                 bias = False,
                 activation = torch.nn.ReLU(),
                 bilinear = True
                ):
        """
        A downscaling component with a double convolution.

        Parameters
        ----------
        input_channels  : int
                          Number of input channels.
        output_channels : int
                          Number of output channels.
        kernel_size     : int
                          Kernel size.
        bias            : bool 
                          Set to True to let convolutional layers have bias term.
        activation      : torch.nn
                          Nonlinear activation layer to be used. If None, uses torch.nn.ReLU().
        bilinear        : bool
                          If set to True, bilinear sampling is used.
        """
        super(upsample_layer, self).__init__()
        if bilinear:
            self.up = torch.nn.Upsample(scale_factor = 2, mode = 'bilinear', align_corners = True)
            self.conv = double_convolution(
                                           input_channels = input_channels + output_channels,
                                           mid_channels = input_channels // 2,
                                           output_channels = output_channels,
                                           kernel_size = kernel_size,
                                           bias = bias,
                                           activation = activation
                                          )
        else:
            self.up = torch.nn.ConvTranspose2d(input_channels , input_channels // 2, kernel_size = 2, stride = 2)
            self.conv = double_convolution(
                                           input_channels = input_channels,
                                           mid_channels = output_channels,
                                           output_channels = output_channels,
                                           kernel_size = kernel_size,
                                           bias = bias,
                                           activation = activation
                                          )


    def forward(self, x1, x2):
        """
        Forward model.

        Parameters
        ----------
        x1             : torch.tensor
                         First input data.
        x2             : torch.tensor
                         Second input data.


        Returns
        ----------
        result        : torch.tensor
                        Result of the forward operation
        """ 
        x1 = self.up(x1)
        diffY = x2.size()[2] - x1.size()[2]
        diffX = x2.size()[3] - x1.size()[3]
        x1 = torch.nn.functional.pad(x1, [diffX // 2, diffX - diffX // 2,
                                          diffY // 2, diffY - diffY // 2])
        x = torch.cat([x2, x1], dim = 1)
        result = self.conv(x)
        return result

__init__(input_channels, output_channels, kernel_size=3, bias=False, activation=torch.nn.ReLU(), bilinear=True)

A downscaling component with a double convolution.

Parameters:

  • input_channels – 
              Number of input channels.
  • output_channels (int) – 
              Number of output channels.
  • kernel_size – 
              Kernel size.
  • bias – 
              Set to True to let convolutional layers have bias term.
  • activation – 
              Nonlinear activation layer to be used. If None, uses torch.nn.ReLU().
  • bilinear – 
              If set to True, bilinear sampling is used.
Source code in odak/learn/models/components.py 
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
def __init__(
             self,
             input_channels,
             output_channels,
             kernel_size = 3,
             bias = False,
             activation = torch.nn.ReLU(),
             bilinear = True
            ):
    """
    A downscaling component with a double convolution.

    Parameters
    ----------
    input_channels  : int
                      Number of input channels.
    output_channels : int
                      Number of output channels.
    kernel_size     : int
                      Kernel size.
    bias            : bool 
                      Set to True to let convolutional layers have bias term.
    activation      : torch.nn
                      Nonlinear activation layer to be used. If None, uses torch.nn.ReLU().
    bilinear        : bool
                      If set to True, bilinear sampling is used.
    """
    super(upsample_layer, self).__init__()
    if bilinear:
        self.up = torch.nn.Upsample(scale_factor = 2, mode = 'bilinear', align_corners = True)
        self.conv = double_convolution(
                                       input_channels = input_channels + output_channels,
                                       mid_channels = input_channels // 2,
                                       output_channels = output_channels,
                                       kernel_size = kernel_size,
                                       bias = bias,
                                       activation = activation
                                      )
    else:
        self.up = torch.nn.ConvTranspose2d(input_channels , input_channels // 2, kernel_size = 2, stride = 2)
        self.conv = double_convolution(
                                       input_channels = input_channels,
                                       mid_channels = output_channels,
                                       output_channels = output_channels,
                                       kernel_size = kernel_size,
                                       bias = bias,
                                       activation = activation
                                      )

forward(x1, x2)

Forward model.

Parameters:

  • x1 – 
             First input data.
  • x2 – 
             Second input data.

Returns:

  • result ( tensor ) –

    Result of the forward operation

Source code in odak/learn/models/components.py 
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
def forward(self, x1, x2):
    """
    Forward model.

    Parameters
    ----------
    x1             : torch.tensor
                     First input data.
    x2             : torch.tensor
                     Second input data.


    Returns
    ----------
    result        : torch.tensor
                    Result of the forward operation
    """ 
    x1 = self.up(x1)
    diffY = x2.size()[2] - x1.size()[2]
    diffX = x2.size()[3] - x1.size()[3]
    x1 = torch.nn.functional.pad(x1, [diffX // 2, diffX - diffX // 2,
                                      diffY // 2, diffY - diffY // 2])
    x = torch.cat([x2, x1], dim = 1)
    result = self.conv(x)
    return result

unet

Bases: Module

A U-Net model, heavily inspired from https://github.com/milesial/Pytorch-UNet/tree/master/unet and more can be read from Ronneberger, Olaf, Philipp Fischer, and Thomas Brox. "U-net: Convolutional networks for biomedical image segmentation." Medical Image Computing and Computer-Assisted Intervention–MICCAI 2015: 18th International Conference, Munich, Germany, October 5-9, 2015, Proceedings, Part III 18. Springer International Publishing, 2015.

Source code in odak/learn/models/models.py 
  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
class unet(torch.nn.Module):
    """
    A U-Net model, heavily inspired from `https://github.com/milesial/Pytorch-UNet/tree/master/unet` and more can be read from Ronneberger, Olaf, Philipp Fischer, and Thomas Brox. "U-net: Convolutional networks for biomedical image segmentation." Medical Image Computing and Computer-Assisted Intervention–MICCAI 2015: 18th International Conference, Munich, Germany, October 5-9, 2015, Proceedings, Part III 18. Springer International Publishing, 2015.
    """

    def __init__(
                 self, 
                 depth = 4,
                 dimensions = 64, 
                 input_channels = 2, 
                 output_channels = 1, 
                 bilinear = False,
                 kernel_size = 3,
                 bias = False,
                 activation = torch.nn.ReLU(inplace = True),
                ):
        """
        U-Net model.

        Parameters
        ----------
        depth             : int
                            Number of upsampling and downsampling
        dimensions        : int
                            Number of dimensions.
        input_channels    : int
                            Number of input channels.
        output_channels   : int
                            Number of output channels.
        bilinear          : bool
                            Uses bilinear upsampling in upsampling layers when set True.
        bias              : bool
                            Set True to let convolutional layers learn a bias term.
        activation        : torch.nn
                            Non-linear activation layer to be used (e.g., torch.nn.ReLU(), torch.nn.Sigmoid().
        """
        super(unet, self).__init__()
        self.inc = double_convolution(
                                      input_channels = input_channels,
                                      mid_channels = dimensions,
                                      output_channels = dimensions,
                                      kernel_size = kernel_size,
                                      bias = bias,
                                      activation = activation
                                     )
        factor = 2 if bilinear else 1        
        self.downsampling_layers = torch.nn.ModuleList()
        self.upsampling_layers = torch.nn.ModuleList()
        for i in range(depth): # downsampling layers
            in_channels = dimensions * (2 ** i)
            out_channels = dimensions * (2 ** (i + 1))
            down_layer = downsample_layer(in_channels,
                                            out_channels,
                                            kernel_size=kernel_size,
                                            bias=bias,
                                            activation=activation
                                            )
            self.downsampling_layers.append(down_layer)      

        for i in range(depth - 1, -1, -1):  # upsampling layers
            up_in_channels = dimensions * (2 ** (i + 1))  
            up_out_channels = dimensions * (2 ** i) 
            up_layer = upsample_layer(up_in_channels, up_out_channels, kernel_size=kernel_size, bias=bias, activation=activation, bilinear=bilinear)
            self.upsampling_layers.append(up_layer)

        self.outc = torch.nn.Conv2d(
                                    dimensions, 
                                    output_channels,
                                    kernel_size = kernel_size,
                                    padding = kernel_size // 2,
                                    bias = bias
                                   )
    def forward(self, x):
        """
        Forward model.

        Parameters
        ----------
        x             : torch.tensor
                        Input data.


        Returns
        ----------
        result        : torch.tensor
                        Estimated output.      
        """
        downsampling_outputs = [self.inc(x)]
        for down_layer in self.downsampling_layers:
            x_down = down_layer(downsampling_outputs[-1])
            downsampling_outputs.append(x_down)
        x_up = downsampling_outputs[-1]
        for i, up_layer in enumerate((self.upsampling_layers)):
            x_up = up_layer(x_up, downsampling_outputs[-(i + 2)])       
        result = self.outc(x_up)
        return result

__init__(depth=4, dimensions=64, input_channels=2, output_channels=1, bilinear=False, kernel_size=3, bias=False, activation=torch.nn.ReLU(inplace=True))

U-Net model.

Parameters:

  • depth – 
                Number of upsampling and downsampling
  • dimensions – 
                Number of dimensions.
  • input_channels – 
                Number of input channels.
  • output_channels – 
                Number of output channels.
  • bilinear – 
                Uses bilinear upsampling in upsampling layers when set True.
  • bias – 
                Set True to let convolutional layers learn a bias term.
  • activation – 
                Non-linear activation layer to be used (e.g., torch.nn.ReLU(), torch.nn.Sigmoid().
Source code in odak/learn/models/models.py 
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
def __init__(
             self, 
             depth = 4,
             dimensions = 64, 
             input_channels = 2, 
             output_channels = 1, 
             bilinear = False,
             kernel_size = 3,
             bias = False,
             activation = torch.nn.ReLU(inplace = True),
            ):
    """
    U-Net model.

    Parameters
    ----------
    depth             : int
                        Number of upsampling and downsampling
    dimensions        : int
                        Number of dimensions.
    input_channels    : int
                        Number of input channels.
    output_channels   : int
                        Number of output channels.
    bilinear          : bool
                        Uses bilinear upsampling in upsampling layers when set True.
    bias              : bool
                        Set True to let convolutional layers learn a bias term.
    activation        : torch.nn
                        Non-linear activation layer to be used (e.g., torch.nn.ReLU(), torch.nn.Sigmoid().
    """
    super(unet, self).__init__()
    self.inc = double_convolution(
                                  input_channels = input_channels,
                                  mid_channels = dimensions,
                                  output_channels = dimensions,
                                  kernel_size = kernel_size,
                                  bias = bias,
                                  activation = activation
                                 )
    factor = 2 if bilinear else 1        
    self.downsampling_layers = torch.nn.ModuleList()
    self.upsampling_layers = torch.nn.ModuleList()
    for i in range(depth): # downsampling layers
        in_channels = dimensions * (2 ** i)
        out_channels = dimensions * (2 ** (i + 1))
        down_layer = downsample_layer(in_channels,
                                        out_channels,
                                        kernel_size=kernel_size,
                                        bias=bias,
                                        activation=activation
                                        )
        self.downsampling_layers.append(down_layer)      

    for i in range(depth - 1, -1, -1):  # upsampling layers
        up_in_channels = dimensions * (2 ** (i + 1))  
        up_out_channels = dimensions * (2 ** i) 
        up_layer = upsample_layer(up_in_channels, up_out_channels, kernel_size=kernel_size, bias=bias, activation=activation, bilinear=bilinear)
        self.upsampling_layers.append(up_layer)

    self.outc = torch.nn.Conv2d(
                                dimensions, 
                                output_channels,
                                kernel_size = kernel_size,
                                padding = kernel_size // 2,
                                bias = bias
                               )

forward(x)

Forward model.

Parameters:

  • x – 
            Input data.

Returns:

  • result ( tensor ) –

    Estimated output.

Source code in odak/learn/models/models.py 
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
def forward(self, x):
    """
    Forward model.

    Parameters
    ----------
    x             : torch.tensor
                    Input data.


    Returns
    ----------
    result        : torch.tensor
                    Estimated output.      
    """
    downsampling_outputs = [self.inc(x)]
    for down_layer in self.downsampling_layers:
        x_down = down_layer(downsampling_outputs[-1])
        downsampling_outputs.append(x_down)
    x_up = downsampling_outputs[-1]
    for i, up_layer in enumerate((self.upsampling_layers)):
        x_up = up_layer(x_up, downsampling_outputs[-(i + 2)])       
    result = self.outc(x_up)
    return result