File size: 13,183 Bytes
b334e29
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import torch
import torch.nn as nn
import torch.utils.checkpoint as cp
from annotator.uniformer.mmcv.cnn import (ConvModule, build_conv_layer, build_norm_layer,
                      constant_init, kaiming_init)
from annotator.uniformer.mmcv.runner import load_checkpoint
from annotator.uniformer.mmcv.utils.parrots_wrapper import _BatchNorm

from annotator.uniformer.mmseg.utils import get_root_logger
from ..builder import BACKBONES


class GlobalContextExtractor(nn.Module):
    """Global Context Extractor for CGNet.

    This class is employed to refine the joint feature of both local feature
    and surrounding context.

    Args:
        channel (int): Number of input feature channels.
        reduction (int): Reductions for global context extractor. Default: 16.
        with_cp (bool): Use checkpoint or not. Using checkpoint will save some
            memory while slowing down the training speed. Default: False.
    """

    def __init__(self, channel, reduction=16, with_cp=False):
        super(GlobalContextExtractor, self).__init__()
        self.channel = channel
        self.reduction = reduction
        assert reduction >= 1 and channel >= reduction
        self.with_cp = with_cp
        self.avg_pool = nn.AdaptiveAvgPool2d(1)
        self.fc = nn.Sequential(
            nn.Linear(channel, channel // reduction), nn.ReLU(inplace=True),
            nn.Linear(channel // reduction, channel), nn.Sigmoid())

    def forward(self, x):

        def _inner_forward(x):
            num_batch, num_channel = x.size()[:2]
            y = self.avg_pool(x).view(num_batch, num_channel)
            y = self.fc(y).view(num_batch, num_channel, 1, 1)
            return x * y

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

        return out


class ContextGuidedBlock(nn.Module):
    """Context Guided Block for CGNet.

    This class consists of four components: local feature extractor,
    surrounding feature extractor, joint feature extractor and global
    context extractor.

    Args:
        in_channels (int): Number of input feature channels.
        out_channels (int): Number of output feature channels.
        dilation (int): Dilation rate for surrounding context extractor.
            Default: 2.
        reduction (int): Reduction for global context extractor. Default: 16.
        skip_connect (bool): Add input to output or not. Default: True.
        downsample (bool): Downsample the input to 1/2 or not. Default: False.
        conv_cfg (dict): Config dict for convolution layer.
            Default: None, which means using conv2d.
        norm_cfg (dict): Config dict for normalization layer.
            Default: dict(type='BN', requires_grad=True).
        act_cfg (dict): Config dict for activation layer.
            Default: dict(type='PReLU').
        with_cp (bool): Use checkpoint or not. Using checkpoint will save some
            memory while slowing down the training speed. Default: False.
    """

    def __init__(self,
                 in_channels,
                 out_channels,
                 dilation=2,
                 reduction=16,
                 skip_connect=True,
                 downsample=False,
                 conv_cfg=None,
                 norm_cfg=dict(type='BN', requires_grad=True),
                 act_cfg=dict(type='PReLU'),
                 with_cp=False):
        super(ContextGuidedBlock, self).__init__()
        self.with_cp = with_cp
        self.downsample = downsample

        channels = out_channels if downsample else out_channels // 2
        if 'type' in act_cfg and act_cfg['type'] == 'PReLU':
            act_cfg['num_parameters'] = channels
        kernel_size = 3 if downsample else 1
        stride = 2 if downsample else 1
        padding = (kernel_size - 1) // 2

        self.conv1x1 = ConvModule(
            in_channels,
            channels,
            kernel_size,
            stride,
            padding,
            conv_cfg=conv_cfg,
            norm_cfg=norm_cfg,
            act_cfg=act_cfg)

        self.f_loc = build_conv_layer(
            conv_cfg,
            channels,
            channels,
            kernel_size=3,
            padding=1,
            groups=channels,
            bias=False)
        self.f_sur = build_conv_layer(
            conv_cfg,
            channels,
            channels,
            kernel_size=3,
            padding=dilation,
            groups=channels,
            dilation=dilation,
            bias=False)

        self.bn = build_norm_layer(norm_cfg, 2 * channels)[1]
        self.activate = nn.PReLU(2 * channels)

        if downsample:
            self.bottleneck = build_conv_layer(
                conv_cfg,
                2 * channels,
                out_channels,
                kernel_size=1,
                bias=False)

        self.skip_connect = skip_connect and not downsample
        self.f_glo = GlobalContextExtractor(out_channels, reduction, with_cp)

    def forward(self, x):

        def _inner_forward(x):
            out = self.conv1x1(x)
            loc = self.f_loc(out)
            sur = self.f_sur(out)

            joi_feat = torch.cat([loc, sur], 1)  # the joint feature
            joi_feat = self.bn(joi_feat)
            joi_feat = self.activate(joi_feat)
            if self.downsample:
                joi_feat = self.bottleneck(joi_feat)  # channel = out_channels
            # f_glo is employed to refine the joint feature
            out = self.f_glo(joi_feat)

            if self.skip_connect:
                return x + out
            else:
                return out

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

        return out


class InputInjection(nn.Module):
    """Downsampling module for CGNet."""

    def __init__(self, num_downsampling):
        super(InputInjection, self).__init__()
        self.pool = nn.ModuleList()
        for i in range(num_downsampling):
            self.pool.append(nn.AvgPool2d(3, stride=2, padding=1))

    def forward(self, x):
        for pool in self.pool:
            x = pool(x)
        return x


@BACKBONES.register_module()
class CGNet(nn.Module):
    """CGNet backbone.

    A Light-weight Context Guided Network for Semantic Segmentation
    arXiv: https://arxiv.org/abs/1811.08201

    Args:
        in_channels (int): Number of input image channels. Normally 3.
        num_channels (tuple[int]): Numbers of feature channels at each stages.
            Default: (32, 64, 128).
        num_blocks (tuple[int]): Numbers of CG blocks at stage 1 and stage 2.
            Default: (3, 21).
        dilations (tuple[int]): Dilation rate for surrounding context
            extractors at stage 1 and stage 2. Default: (2, 4).
        reductions (tuple[int]): Reductions for global context extractors at
            stage 1 and stage 2. Default: (8, 16).
        conv_cfg (dict): Config dict for convolution layer.
            Default: None, which means using conv2d.
        norm_cfg (dict): Config dict for normalization layer.
            Default: dict(type='BN', requires_grad=True).
        act_cfg (dict): Config dict for activation layer.
            Default: dict(type='PReLU').
        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.
        with_cp (bool): Use checkpoint or not. Using checkpoint will save some
            memory while slowing down the training speed. Default: False.
    """

    def __init__(self,
                 in_channels=3,
                 num_channels=(32, 64, 128),
                 num_blocks=(3, 21),
                 dilations=(2, 4),
                 reductions=(8, 16),
                 conv_cfg=None,
                 norm_cfg=dict(type='BN', requires_grad=True),
                 act_cfg=dict(type='PReLU'),
                 norm_eval=False,
                 with_cp=False):

        super(CGNet, self).__init__()
        self.in_channels = in_channels
        self.num_channels = num_channels
        assert isinstance(self.num_channels, tuple) and len(
            self.num_channels) == 3
        self.num_blocks = num_blocks
        assert isinstance(self.num_blocks, tuple) and len(self.num_blocks) == 2
        self.dilations = dilations
        assert isinstance(self.dilations, tuple) and len(self.dilations) == 2
        self.reductions = reductions
        assert isinstance(self.reductions, tuple) and len(self.reductions) == 2
        self.conv_cfg = conv_cfg
        self.norm_cfg = norm_cfg
        self.act_cfg = act_cfg
        if 'type' in self.act_cfg and self.act_cfg['type'] == 'PReLU':
            self.act_cfg['num_parameters'] = num_channels[0]
        self.norm_eval = norm_eval
        self.with_cp = with_cp

        cur_channels = in_channels
        self.stem = nn.ModuleList()
        for i in range(3):
            self.stem.append(
                ConvModule(
                    cur_channels,
                    num_channels[0],
                    3,
                    2 if i == 0 else 1,
                    padding=1,
                    conv_cfg=conv_cfg,
                    norm_cfg=norm_cfg,
                    act_cfg=act_cfg))
            cur_channels = num_channels[0]

        self.inject_2x = InputInjection(1)  # down-sample for Input, factor=2
        self.inject_4x = InputInjection(2)  # down-sample for Input, factor=4

        cur_channels += in_channels
        self.norm_prelu_0 = nn.Sequential(
            build_norm_layer(norm_cfg, cur_channels)[1],
            nn.PReLU(cur_channels))

        # stage 1
        self.level1 = nn.ModuleList()
        for i in range(num_blocks[0]):
            self.level1.append(
                ContextGuidedBlock(
                    cur_channels if i == 0 else num_channels[1],
                    num_channels[1],
                    dilations[0],
                    reductions[0],
                    downsample=(i == 0),
                    conv_cfg=conv_cfg,
                    norm_cfg=norm_cfg,
                    act_cfg=act_cfg,
                    with_cp=with_cp))  # CG block

        cur_channels = 2 * num_channels[1] + in_channels
        self.norm_prelu_1 = nn.Sequential(
            build_norm_layer(norm_cfg, cur_channels)[1],
            nn.PReLU(cur_channels))

        # stage 2
        self.level2 = nn.ModuleList()
        for i in range(num_blocks[1]):
            self.level2.append(
                ContextGuidedBlock(
                    cur_channels if i == 0 else num_channels[2],
                    num_channels[2],
                    dilations[1],
                    reductions[1],
                    downsample=(i == 0),
                    conv_cfg=conv_cfg,
                    norm_cfg=norm_cfg,
                    act_cfg=act_cfg,
                    with_cp=with_cp))  # CG block

        cur_channels = 2 * num_channels[2]
        self.norm_prelu_2 = nn.Sequential(
            build_norm_layer(norm_cfg, cur_channels)[1],
            nn.PReLU(cur_channels))

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

        # stage 0
        inp_2x = self.inject_2x(x)
        inp_4x = self.inject_4x(x)
        for layer in self.stem:
            x = layer(x)
        x = self.norm_prelu_0(torch.cat([x, inp_2x], 1))
        output.append(x)

        # stage 1
        for i, layer in enumerate(self.level1):
            x = layer(x)
            if i == 0:
                down1 = x
        x = self.norm_prelu_1(torch.cat([x, down1, inp_4x], 1))
        output.append(x)

        # stage 2
        for i, layer in enumerate(self.level2):
            x = layer(x)
            if i == 0:
                down2 = x
        x = self.norm_prelu_2(torch.cat([down2, x], 1))
        output.append(x)

        return output

    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, nn.Linear)):
                    kaiming_init(m)
                elif isinstance(m, (_BatchNorm, nn.GroupNorm)):
                    constant_init(m, 1)
                elif isinstance(m, nn.PReLU):
                    constant_init(m, 0)
        else:
            raise TypeError('pretrained must be a str or None')

    def train(self, mode=True):
        """Convert the model into training mode will keeping the normalization
        layer freezed."""
        super(CGNet, self).train(mode)
        if mode and self.norm_eval:
            for m in self.modules():
                # trick: eval have effect on BatchNorm only
                if isinstance(m, _BatchNorm):
                    m.eval()