File size: 29,836 Bytes
24be7a2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
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
198
199
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
241
242
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
329
330
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
404
405
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
465
466
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
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
import torch
import torch.nn as nn
import torch.utils.checkpoint as cp
from mmcv.cnn import (UPSAMPLE_LAYERS, ConvModule, build_activation_layer,
                      build_norm_layer, build_upsample_layer, constant_init,
                      kaiming_init)
from mmcv.runner import load_checkpoint
from mmcv.utils.parrots_wrapper import _BatchNorm
from mmseg.utils import get_root_logger


class UpConvBlock(nn.Module):
    """Upsample convolution block in decoder for UNet.

    This upsample convolution block consists of one upsample module
    followed by one convolution block. The upsample module expands the
    high-level low-resolution feature map and the convolution block fuses
    the upsampled high-level low-resolution feature map and the low-level
    high-resolution feature map from encoder.

    Args:
        conv_block (nn.Sequential): Sequential of convolutional layers.
        in_channels (int): Number of input channels of the high-level
        skip_channels (int): Number of input channels of the low-level
        high-resolution feature map from encoder.
        out_channels (int): Number of output channels.
        num_convs (int): Number of convolutional layers in the conv_block.
            Default: 2.
        stride (int): Stride of convolutional layer in conv_block. Default: 1.
        dilation (int): Dilation rate of convolutional layer in conv_block.
            Default: 1.
        with_cp (bool): Use checkpoint or not. Using checkpoint will save some
            memory while slowing down the training speed. Default: False.
        conv_cfg (dict | None): Config dict for convolution layer.
            Default: None.
        norm_cfg (dict | None): Config dict for normalization layer.
            Default: dict(type='BN').
        act_cfg (dict | None): Config dict for activation layer in ConvModule.
            Default: dict(type='ReLU').
        upsample_cfg (dict): The upsample config of the upsample module in
            decoder. Default: dict(type='InterpConv'). If the size of
            high-level feature map is the same as that of skip feature map
            (low-level feature map from encoder), it does not need upsample the
            high-level feature map and the upsample_cfg is None.
        dcn (bool): Use deformable convoluton in convolutional layer or not.
            Default: None.
        plugins (dict): plugins for convolutional layers. Default: None.
    """

    def __init__(self,
                 conv_block,
                 in_channels,
                 skip_channels,
                 out_channels,
                 num_convs=2,
                 stride=1,
                 dilation=1,
                 with_cp=False,
                 conv_cfg=None,
                 norm_cfg=dict(type='BN'),
                 act_cfg=dict(type='ReLU'),
                 upsample_cfg=dict(type='InterpConv'),
                 dcn=None,
                 plugins=None):
        super(UpConvBlock, self).__init__()
        assert dcn is None, 'Not implemented yet.'
        assert plugins is None, 'Not implemented yet.'

        self.conv_block = conv_block(
            in_channels=2 * skip_channels,
            out_channels=out_channels,
            num_convs=num_convs,
            stride=stride,
            dilation=dilation,
            with_cp=with_cp,
            conv_cfg=conv_cfg,
            norm_cfg=norm_cfg,
            act_cfg=act_cfg,
            dcn=None,
            plugins=None)
        if upsample_cfg is not None:
            self.upsample = build_upsample_layer(
                cfg=upsample_cfg,
                in_channels=in_channels,
                out_channels=skip_channels,
                with_cp=with_cp,
                norm_cfg=norm_cfg,
                act_cfg=act_cfg)
        else:
            self.upsample = ConvModule(
                in_channels,
                skip_channels,
                kernel_size=1,
                stride=1,
                padding=0,
                conv_cfg=conv_cfg,
                norm_cfg=norm_cfg,
                act_cfg=act_cfg)

    def forward(self, skip, x):
        """Forward function."""

        x = self.upsample(x)
        out = torch.cat([skip, x], dim=1)
        out = self.conv_block(out)

        return out


class BasicConvBlock(nn.Module):
    """Basic convolutional block for UNet.

    This module consists of several plain convolutional layers.

    Args:
        in_channels (int): Number of input channels.
        out_channels (int): Number of output channels.
        num_convs (int): Number of convolutional layers. Default: 2.
        stride (int): Whether use stride convolution to downsample
            the input feature map. If stride=2, it only uses stride convolution
            in the first convolutional layer to downsample the input feature
            map. Options are 1 or 2. Default: 1.
        dilation (int): Whether use dilated convolution to expand the
            receptive field. Set dilation rate of each convolutional layer and
            the dilation rate of the first convolutional layer is always 1.
            Default: 1.
        with_cp (bool): Use checkpoint or not. Using checkpoint will save some
            memory while slowing down the training speed. Default: False.
        conv_cfg (dict | None): Config dict for convolution layer.
            Default: None.
        norm_cfg (dict | None): Config dict for normalization layer.
            Default: dict(type='BN').
        act_cfg (dict | None): Config dict for activation layer in ConvModule.
            Default: dict(type='ReLU').
        dcn (bool): Use deformable convoluton in convolutional layer or not.
            Default: None.
        plugins (dict): plugins for convolutional layers. Default: None.
    """

    def __init__(self,
                 in_channels,
                 out_channels,
                 num_convs=2,
                 stride=1,
                 dilation=1,
                 with_cp=False,
                 conv_cfg=None,
                 norm_cfg=dict(type='BN'),
                 act_cfg=dict(type='ReLU'),
                 dcn=None,
                 plugins=None):
        super(BasicConvBlock, self).__init__()
        assert dcn is None, 'Not implemented yet.'
        assert plugins is None, 'Not implemented yet.'

        self.with_cp = with_cp
        convs = []
        for i in range(num_convs):
            convs.append(
                ConvModule(
                    in_channels=in_channels if i == 0 else out_channels,
                    out_channels=out_channels,
                    kernel_size=3,
                    stride=stride if i == 0 else 1,
                    dilation=1 if i == 0 else dilation,
                    padding=1 if i == 0 else dilation,
                    conv_cfg=conv_cfg,
                    norm_cfg=norm_cfg,
                    act_cfg=act_cfg))

        self.convs = nn.Sequential(*convs)

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

        if self.with_cp and x.requires_grad:
            out = cp.checkpoint(self.convs, x)
        else:
            out = self.convs(x)
        return out


class DeconvModule(nn.Module):
    """Deconvolution upsample module in decoder for UNet (2X upsample).

    This module uses deconvolution to upsample feature map in the decoder
    of UNet.

    Args:
        in_channels (int): Number of input channels.
        out_channels (int): Number of output channels.
        with_cp (bool): Use checkpoint or not. Using checkpoint will save some
            memory while slowing down the training speed. Default: False.
        norm_cfg (dict | None): Config dict for normalization layer.
            Default: dict(type='BN').
        act_cfg (dict | None): Config dict for activation layer in ConvModule.
            Default: dict(type='ReLU').
        kernel_size (int): Kernel size of the convolutional layer. Default: 4.
    """

    def __init__(self,
                 in_channels,
                 out_channels,
                 with_cp=False,
                 norm_cfg=dict(type='BN'),
                 act_cfg=dict(type='ReLU'),
                 *,
                 kernel_size=4,
                 scale_factor=2):
        super(DeconvModule, self).__init__()

        assert (kernel_size - scale_factor >= 0) and\
               (kernel_size - scale_factor) % 2 == 0,\
               f'kernel_size should be greater than or equal to scale_factor '\
               f'and (kernel_size - scale_factor) should be even numbers, '\
               f'while the kernel size is {kernel_size} and scale_factor is '\
               f'{scale_factor}.'

        stride = scale_factor
        padding = (kernel_size - scale_factor) // 2
        self.with_cp = with_cp
        deconv = nn.ConvTranspose2d(
            in_channels,
            out_channels,
            kernel_size=kernel_size,
            stride=stride,
            padding=padding)

        norm_name, norm = build_norm_layer(norm_cfg, out_channels)
        activate = build_activation_layer(act_cfg)
        self.deconv_upsamping = nn.Sequential(deconv, norm, activate)

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

        if self.with_cp and x.requires_grad:
            out = cp.checkpoint(self.deconv_upsamping, x)
        else:
            out = self.deconv_upsamping(x)
        return out


@UPSAMPLE_LAYERS.register_module()
class InterpConv(nn.Module):
    """Interpolation upsample module in decoder for UNet.

    This module uses interpolation to upsample feature map in the decoder
    of UNet. It consists of one interpolation upsample layer and one
    convolutional layer. It can be one interpolation upsample layer followed
    by one convolutional layer (conv_first=False) or one convolutional layer
    followed by one interpolation upsample layer (conv_first=True).

    Args:
        in_channels (int): Number of input channels.
        out_channels (int): Number of output channels.
        with_cp (bool): Use checkpoint or not. Using checkpoint will save some
            memory while slowing down the training speed. Default: False.
        norm_cfg (dict | None): Config dict for normalization layer.
            Default: dict(type='BN').
        act_cfg (dict | None): Config dict for activation layer in ConvModule.
            Default: dict(type='ReLU').
        conv_cfg (dict | None): Config dict for convolution layer.
            Default: None.
        conv_first (bool): Whether convolutional layer or interpolation
            upsample layer first. Default: False. It means interpolation
            upsample layer followed by one convolutional layer.
        kernel_size (int): Kernel size of the convolutional layer. Default: 1.
        stride (int): Stride of the convolutional layer. Default: 1.
        padding (int): Padding of the convolutional layer. Default: 1.
        upsampe_cfg (dict): Interpolation config of the upsample layer.
            Default: dict(
                scale_factor=2, mode='bilinear', align_corners=False).
    """

    def __init__(self,
                 in_channels,
                 out_channels,
                 with_cp=False,
                 norm_cfg=dict(type='BN'),
                 act_cfg=dict(type='ReLU'),
                 *,
                 conv_cfg=None,
                 conv_first=False,
                 kernel_size=1,
                 stride=1,
                 padding=0,
                 upsampe_cfg=dict(
                     scale_factor=2, mode='bilinear', align_corners=False)):
        super(InterpConv, self).__init__()

        self.with_cp = with_cp
        conv = ConvModule(
            in_channels,
            out_channels,
            kernel_size=kernel_size,
            stride=stride,
            padding=padding,
            conv_cfg=conv_cfg,
            norm_cfg=norm_cfg,
            act_cfg=act_cfg)
        upsample = nn.Upsample(**upsampe_cfg)
        if conv_first:
            self.interp_upsample = nn.Sequential(conv, upsample)
        else:
            self.interp_upsample = nn.Sequential(upsample, conv)

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

        if self.with_cp and x.requires_grad:
            out = cp.checkpoint(self.interp_upsample, x)
        else:
            out = self.interp_upsample(x)
        return out


class UNet(nn.Module):
    """UNet backbone.
    U-Net: Convolutional Networks for Biomedical Image Segmentation.
    https://arxiv.org/pdf/1505.04597.pdf

    Args:
        in_channels (int): Number of input image channels. Default" 3.
        base_channels (int): Number of base channels of each stage.
            The output channels of the first stage. Default: 64.
        num_stages (int): Number of stages in encoder, normally 5. Default: 5.
        strides (Sequence[int 1 | 2]): Strides of each stage in encoder.
            len(strides) is equal to num_stages. Normally the stride of the
            first stage in encoder is 1. If strides[i]=2, it uses stride
            convolution to downsample in the correspondence encoder stage.
            Default: (1, 1, 1, 1, 1).
        enc_num_convs (Sequence[int]): Number of convolutional layers in the
            convolution block of the correspondence encoder stage.
            Default: (2, 2, 2, 2, 2).
        dec_num_convs (Sequence[int]): Number of convolutional layers in the
            convolution block of the correspondence decoder stage.
            Default: (2, 2, 2, 2).
        downsamples (Sequence[int]): Whether use MaxPool to downsample the
            feature map after the first stage of encoder
            (stages: [1, num_stages)). If the correspondence encoder stage use
            stride convolution (strides[i]=2), it will never use MaxPool to
            downsample, even downsamples[i-1]=True.
            Default: (True, True, True, True).
        enc_dilations (Sequence[int]): Dilation rate of each stage in encoder.
            Default: (1, 1, 1, 1, 1).
        dec_dilations (Sequence[int]): Dilation rate of each stage in decoder.
            Default: (1, 1, 1, 1).
        with_cp (bool): Use checkpoint or not. Using checkpoint will save some
            memory while slowing down the training speed. Default: False.
        conv_cfg (dict | None): Config dict for convolution layer.
            Default: None.
        norm_cfg (dict | None): Config dict for normalization layer.
            Default: dict(type='BN').
        act_cfg (dict | None): Config dict for activation layer in ConvModule.
            Default: dict(type='ReLU').
        upsample_cfg (dict): The upsample config of the upsample module in
            decoder. Default: dict(type='InterpConv').
        norm_eval (bool): Whether to set norm layers to eval mode, namely,
            freeze running stats (mean and var). Note: Effect on Batch Norm
            and its variants only. Default: False.
        dcn (bool): Use deformable convolution in convolutional layer or not.
            Default: None.
        plugins (dict): plugins for convolutional layers. Default: None.

    Notice:
        The input image size should be devisible by the whole downsample rate
        of the encoder. More detail of the whole downsample rate can be found
        in UNet._check_input_devisible.

    """

    def __init__(self,
                 in_channels=3,
                 base_channels=64,
                 num_stages=5,
                 strides=(1, 1, 1, 1, 1),
                 enc_num_convs=(2, 2, 2, 2, 2),
                 dec_num_convs=(2, 2, 2, 2),
                 downsamples=(True, True, True, True),
                 enc_dilations=(1, 1, 1, 1, 1),
                 dec_dilations=(1, 1, 1, 1),
                 with_cp=False,
                 conv_cfg=None,
                 norm_cfg=dict(type='BN'),
                 act_cfg=dict(type='ReLU'),
                 upsample_cfg=dict(type='InterpConv'),
                 norm_eval=False,
                 dcn=None,
                 plugins=None):
        super(UNet, self).__init__()
        assert dcn is None, 'Not implemented yet.'
        assert plugins is None, 'Not implemented yet.'
        assert len(strides) == num_stages, \
            'The length of strides should be equal to num_stages, '\
            f'while the strides is {strides}, the length of '\
            f'strides is {len(strides)}, and the num_stages is '\
            f'{num_stages}.'
        assert len(enc_num_convs) == num_stages, \
            'The length of enc_num_convs should be equal to num_stages, '\
            f'while the enc_num_convs is {enc_num_convs}, the length of '\
            f'enc_num_convs is {len(enc_num_convs)}, and the num_stages is '\
            f'{num_stages}.'
        assert len(dec_num_convs) == (num_stages-1), \
            'The length of dec_num_convs should be equal to (num_stages-1), '\
            f'while the dec_num_convs is {dec_num_convs}, the length of '\
            f'dec_num_convs is {len(dec_num_convs)}, and the num_stages is '\
            f'{num_stages}.'
        assert len(downsamples) == (num_stages-1), \
            'The length of downsamples should be equal to (num_stages-1), '\
            f'while the downsamples is {downsamples}, the length of '\
            f'downsamples is {len(downsamples)}, and the num_stages is '\
            f'{num_stages}.'
        assert len(enc_dilations) == num_stages, \
            'The length of enc_dilations should be equal to num_stages, '\
            f'while the enc_dilations is {enc_dilations}, the length of '\
            f'enc_dilations is {len(enc_dilations)}, and the num_stages is '\
            f'{num_stages}.'
        assert len(dec_dilations) == (num_stages-1), \
            'The length of dec_dilations should be equal to (num_stages-1), '\
            f'while the dec_dilations is {dec_dilations}, the length of '\
            f'dec_dilations is {len(dec_dilations)}, and the num_stages is '\
            f'{num_stages}.'
        self.num_stages = num_stages
        self.strides = strides
        self.downsamples = downsamples
        self.norm_eval = norm_eval

        self.encoder = nn.ModuleList()
        self.decoder = nn.ModuleList()

        for i in range(num_stages):
            enc_conv_block = []
            if i != 0:
                if strides[i] == 1 and downsamples[i - 1]:
                    enc_conv_block.append(nn.MaxPool2d(kernel_size=2))
                upsample = (strides[i] != 1 or downsamples[i - 1])
                self.decoder.append(
                    UpConvBlock(
                        conv_block=BasicConvBlock,
                        in_channels=base_channels * 2**i,
                        skip_channels=base_channels * 2**(i - 1),
                        out_channels=base_channels * 2**(i - 1),
                        num_convs=dec_num_convs[i - 1],
                        stride=1,
                        dilation=dec_dilations[i - 1],
                        with_cp=with_cp,
                        conv_cfg=conv_cfg,
                        norm_cfg=norm_cfg,
                        act_cfg=act_cfg,
                        upsample_cfg=upsample_cfg if upsample else None,
                        dcn=None,
                        plugins=None))

            enc_conv_block.append(
                BasicConvBlock(
                    in_channels=in_channels,
                    out_channels=base_channels * 2**i,
                    num_convs=enc_num_convs[i],
                    stride=strides[i],
                    dilation=enc_dilations[i],
                    with_cp=with_cp,
                    conv_cfg=conv_cfg,
                    norm_cfg=norm_cfg,
                    act_cfg=act_cfg,
                    dcn=None,
                    plugins=None))
            self.encoder.append((nn.Sequential(*enc_conv_block)))
            in_channels = base_channels * 2**i

    def forward(self, x):
        enc_outs = []

        for enc in self.encoder:
            x = enc(x)
            enc_outs.append(x)
        dec_outs = [x]
        for i in reversed(range(len(self.decoder))):
            x = self.decoder[i](enc_outs[i], x)
            dec_outs.append(x)

        return dec_outs

    def init_weights(self, pretrained=None):
        """Initialize the weights in backbone.

        Args:
            pretrained (str, optional): Path to pre-trained weights.
                Defaults to None.
        """
        if isinstance(pretrained, str):
            logger = get_root_logger()
            load_checkpoint(self, pretrained, strict=False, logger=logger)
        elif pretrained is None:
            for m in self.modules():
                if isinstance(m, nn.Conv2d):
                    kaiming_init(m)
                elif isinstance(m, (_BatchNorm, nn.GroupNorm)):
                    constant_init(m, 1)
        else:
            raise TypeError('pretrained must be a str or None')


class ShapeUNet(nn.Module):
    """ShapeUNet backbone with small modifications.
    U-Net: Convolutional Networks for Biomedical Image Segmentation.
    https://arxiv.org/pdf/1505.04597.pdf

    Args:
        in_channels (int): Number of input image channels. Default" 3.
        base_channels (int): Number of base channels of each stage.
            The output channels of the first stage. Default: 64.
        num_stages (int): Number of stages in encoder, normally 5. Default: 5.
        strides (Sequence[int 1 | 2]): Strides of each stage in encoder.
            len(strides) is equal to num_stages. Normally the stride of the
            first stage in encoder is 1. If strides[i]=2, it uses stride
            convolution to downsample in the correspondance encoder stage.
            Default: (1, 1, 1, 1, 1).
        enc_num_convs (Sequence[int]): Number of convolutional layers in the
            convolution block of the correspondance encoder stage.
            Default: (2, 2, 2, 2, 2).
        dec_num_convs (Sequence[int]): Number of convolutional layers in the
            convolution block of the correspondance decoder stage.
            Default: (2, 2, 2, 2).
        downsamples (Sequence[int]): Whether use MaxPool to downsample the
            feature map after the first stage of encoder
            (stages: [1, num_stages)). If the correspondance encoder stage use
            stride convolution (strides[i]=2), it will never use MaxPool to
            downsample, even downsamples[i-1]=True.
            Default: (True, True, True, True).
        enc_dilations (Sequence[int]): Dilation rate of each stage in encoder.
            Default: (1, 1, 1, 1, 1).
        dec_dilations (Sequence[int]): Dilation rate of each stage in decoder.
            Default: (1, 1, 1, 1).
        with_cp (bool): Use checkpoint or not. Using checkpoint will save some
            memory while slowing down the training speed. Default: False.
        conv_cfg (dict | None): Config dict for convolution layer.
            Default: None.
        norm_cfg (dict | None): Config dict for normalization layer.
            Default: dict(type='BN').
        act_cfg (dict | None): Config dict for activation layer in ConvModule.
            Default: dict(type='ReLU').
        upsample_cfg (dict): The upsample config of the upsample module in
            decoder. Default: dict(type='InterpConv').
        norm_eval (bool): Whether to set norm layers to eval mode, namely,
            freeze running stats (mean and var). Note: Effect on Batch Norm
            and its variants only. Default: False.
        dcn (bool): Use deformable convoluton in convolutional layer or not.
            Default: None.
        plugins (dict): plugins for convolutional layers. Default: None.

    Notice:
        The input image size should be devisible by the whole downsample rate
        of the encoder. More detail of the whole downsample rate can be found
        in UNet._check_input_devisible.

    """

    def __init__(self,
                 in_channels=3,
                 base_channels=64,
                 num_stages=5,
                 attr_embedding=128,
                 strides=(1, 1, 1, 1, 1),
                 enc_num_convs=(2, 2, 2, 2, 2),
                 dec_num_convs=(2, 2, 2, 2),
                 downsamples=(True, True, True, True),
                 enc_dilations=(1, 1, 1, 1, 1),
                 dec_dilations=(1, 1, 1, 1),
                 with_cp=False,
                 conv_cfg=None,
                 norm_cfg=dict(type='BN'),
                 act_cfg=dict(type='ReLU'),
                 upsample_cfg=dict(type='InterpConv'),
                 norm_eval=False,
                 dcn=None,
                 plugins=None):
        super(ShapeUNet, self).__init__()
        assert dcn is None, 'Not implemented yet.'
        assert plugins is None, 'Not implemented yet.'
        assert len(strides) == num_stages, \
            'The length of strides should be equal to num_stages, '\
            f'while the strides is {strides}, the length of '\
            f'strides is {len(strides)}, and the num_stages is '\
            f'{num_stages}.'
        assert len(enc_num_convs) == num_stages, \
            'The length of enc_num_convs should be equal to num_stages, '\
            f'while the enc_num_convs is {enc_num_convs}, the length of '\
            f'enc_num_convs is {len(enc_num_convs)}, and the num_stages is '\
            f'{num_stages}.'
        assert len(dec_num_convs) == (num_stages-1), \
            'The length of dec_num_convs should be equal to (num_stages-1), '\
            f'while the dec_num_convs is {dec_num_convs}, the length of '\
            f'dec_num_convs is {len(dec_num_convs)}, and the num_stages is '\
            f'{num_stages}.'
        assert len(downsamples) == (num_stages-1), \
            'The length of downsamples should be equal to (num_stages-1), '\
            f'while the downsamples is {downsamples}, the length of '\
            f'downsamples is {len(downsamples)}, and the num_stages is '\
            f'{num_stages}.'
        assert len(enc_dilations) == num_stages, \
            'The length of enc_dilations should be equal to num_stages, '\
            f'while the enc_dilations is {enc_dilations}, the length of '\
            f'enc_dilations is {len(enc_dilations)}, and the num_stages is '\
            f'{num_stages}.'
        assert len(dec_dilations) == (num_stages-1), \
            'The length of dec_dilations should be equal to (num_stages-1), '\
            f'while the dec_dilations is {dec_dilations}, the length of '\
            f'dec_dilations is {len(dec_dilations)}, and the num_stages is '\
            f'{num_stages}.'
        self.num_stages = num_stages
        self.strides = strides
        self.downsamples = downsamples
        self.norm_eval = norm_eval

        self.encoder = nn.ModuleList()
        self.decoder = nn.ModuleList()

        for i in range(num_stages):
            enc_conv_block = []
            if i != 0:
                if strides[i] == 1 and downsamples[i - 1]:
                    enc_conv_block.append(nn.MaxPool2d(kernel_size=2))
                upsample = (strides[i] != 1 or downsamples[i - 1])
                self.decoder.append(
                    UpConvBlock(
                        conv_block=BasicConvBlock,
                        in_channels=base_channels * 2**i,
                        skip_channels=base_channels * 2**(i - 1),
                        out_channels=base_channels * 2**(i - 1),
                        num_convs=dec_num_convs[i - 1],
                        stride=1,
                        dilation=dec_dilations[i - 1],
                        with_cp=with_cp,
                        conv_cfg=conv_cfg,
                        norm_cfg=norm_cfg,
                        act_cfg=act_cfg,
                        upsample_cfg=upsample_cfg if upsample else None,
                        dcn=None,
                        plugins=None))

            enc_conv_block.append(
                BasicConvBlock(
                    in_channels=in_channels + attr_embedding,
                    out_channels=base_channels * 2**i,
                    num_convs=enc_num_convs[i],
                    stride=strides[i],
                    dilation=enc_dilations[i],
                    with_cp=with_cp,
                    conv_cfg=conv_cfg,
                    norm_cfg=norm_cfg,
                    act_cfg=act_cfg,
                    dcn=None,
                    plugins=None))
            self.encoder.append((nn.Sequential(*enc_conv_block)))
            in_channels = base_channels * 2**i

    def forward(self, x, attr_embedding):
        enc_outs = []
        Be, Ce = attr_embedding.size()
        for enc in self.encoder:
            _, _, H, W = x.size()
            x = enc(
                torch.cat([
                    x,
                    attr_embedding.view(Be, Ce, 1, 1).expand((Be, Ce, H, W))
                ],
                          dim=1))
            enc_outs.append(x)
        dec_outs = [x]
        for i in reversed(range(len(self.decoder))):
            x = self.decoder[i](enc_outs[i], x)
            dec_outs.append(x)

        return dec_outs

    def init_weights(self, pretrained=None):
        """Initialize the weights in backbone.

        Args:
            pretrained (str, optional): Path to pre-trained weights.
                Defaults to None.
        """
        if isinstance(pretrained, str):
            logger = get_root_logger()
            load_checkpoint(self, pretrained, strict=False, logger=logger)
        elif pretrained is None:
            for m in self.modules():
                if isinstance(m, nn.Conv2d):
                    kaiming_init(m)
                elif isinstance(m, (_BatchNorm, nn.GroupNorm)):
                    constant_init(m, 1)
        else:
            raise TypeError('pretrained must be a str or None')