Image Segmentation
Transformers
PyTorch
upernet
Inference Endpoints
File size: 10,090 Bytes
b13b124
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import math

import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.utils.checkpoint as cp
from mmcv.cnn import build_conv_layer, build_norm_layer

from ..builder import BACKBONES
from ..utils import ResLayer
from .resnet import Bottleneck as _Bottleneck
from .resnet import ResNetV1d


class RSoftmax(nn.Module):
    """Radix Softmax module in ``SplitAttentionConv2d``.

    Args:
        radix (int): Radix of input.
        groups (int): Groups of input.
    """

    def __init__(self, radix, groups):
        super().__init__()
        self.radix = radix
        self.groups = groups

    def forward(self, x):
        batch = x.size(0)
        if self.radix > 1:
            x = x.view(batch, self.groups, self.radix, -1).transpose(1, 2)
            x = F.softmax(x, dim=1)
            x = x.reshape(batch, -1)
        else:
            x = torch.sigmoid(x)
        return x


class SplitAttentionConv2d(nn.Module):
    """Split-Attention Conv2d in ResNeSt.

    Args:
        in_channels (int): Same as nn.Conv2d.
        out_channels (int): Same as nn.Conv2d.
        kernel_size (int | tuple[int]): Same as nn.Conv2d.
        stride (int | tuple[int]): Same as nn.Conv2d.
        padding (int | tuple[int]): Same as nn.Conv2d.
        dilation (int | tuple[int]): Same as nn.Conv2d.
        groups (int): Same as nn.Conv2d.
        radix (int): Radix of SpltAtConv2d. Default: 2
        reduction_factor (int): Reduction factor of inter_channels. Default: 4.
        conv_cfg (dict): Config dict for convolution layer. Default: None,
            which means using conv2d.
        norm_cfg (dict): Config dict for normalization layer. Default: None.
        dcn (dict): Config dict for DCN. Default: None.
    """

    def __init__(self,
                 in_channels,
                 channels,
                 kernel_size,
                 stride=1,
                 padding=0,
                 dilation=1,
                 groups=1,
                 radix=2,
                 reduction_factor=4,
                 conv_cfg=None,
                 norm_cfg=dict(type='BN'),
                 dcn=None):
        super(SplitAttentionConv2d, self).__init__()
        inter_channels = max(in_channels * radix // reduction_factor, 32)
        self.radix = radix
        self.groups = groups
        self.channels = channels
        self.with_dcn = dcn is not None
        self.dcn = dcn
        fallback_on_stride = False
        if self.with_dcn:
            fallback_on_stride = self.dcn.pop('fallback_on_stride', False)
        if self.with_dcn and not fallback_on_stride:
            assert conv_cfg is None, 'conv_cfg must be None for DCN'
            conv_cfg = dcn
        self.conv = build_conv_layer(
            conv_cfg,
            in_channels,
            channels * radix,
            kernel_size,
            stride=stride,
            padding=padding,
            dilation=dilation,
            groups=groups * radix,
            bias=False)
        self.norm0_name, norm0 = build_norm_layer(
            norm_cfg, channels * radix, postfix=0)
        self.add_module(self.norm0_name, norm0)
        self.relu = nn.ReLU(inplace=True)
        self.fc1 = build_conv_layer(
            None, channels, inter_channels, 1, groups=self.groups)
        self.norm1_name, norm1 = build_norm_layer(
            norm_cfg, inter_channels, postfix=1)
        self.add_module(self.norm1_name, norm1)
        self.fc2 = build_conv_layer(
            None, inter_channels, channels * radix, 1, groups=self.groups)
        self.rsoftmax = RSoftmax(radix, groups)

    @property
    def norm0(self):
        """nn.Module: the normalization layer named "norm0" """
        return getattr(self, self.norm0_name)

    @property
    def norm1(self):
        """nn.Module: the normalization layer named "norm1" """
        return getattr(self, self.norm1_name)

    def forward(self, x):
        x = self.conv(x)
        x = self.norm0(x)
        x = self.relu(x)

        batch, rchannel = x.shape[:2]
        batch = x.size(0)
        if self.radix > 1:
            splits = x.view(batch, self.radix, -1, *x.shape[2:])
            gap = splits.sum(dim=1)
        else:
            gap = x
        gap = F.adaptive_avg_pool2d(gap, 1)
        gap = self.fc1(gap)

        gap = self.norm1(gap)
        gap = self.relu(gap)

        atten = self.fc2(gap)
        atten = self.rsoftmax(atten).view(batch, -1, 1, 1)

        if self.radix > 1:
            attens = atten.view(batch, self.radix, -1, *atten.shape[2:])
            out = torch.sum(attens * splits, dim=1)
        else:
            out = atten * x
        return out.contiguous()


class Bottleneck(_Bottleneck):
    """Bottleneck block for ResNeSt.

    Args:
        inplane (int): Input planes of this block.
        planes (int): Middle planes of this block.
        groups (int): Groups of conv2.
        width_per_group (int): Width per group of conv2. 64x4d indicates
            ``groups=64, width_per_group=4`` and 32x8d indicates
            ``groups=32, width_per_group=8``.
        radix (int): Radix of SpltAtConv2d. Default: 2
        reduction_factor (int): Reduction factor of inter_channels in
            SplitAttentionConv2d. Default: 4.
        avg_down_stride (bool): Whether to use average pool for stride in
            Bottleneck. Default: True.
        kwargs (dict): Key word arguments for base class.
    """
    expansion = 4

    def __init__(self,
                 inplanes,
                 planes,
                 groups=1,
                 base_width=4,
                 base_channels=64,
                 radix=2,
                 reduction_factor=4,
                 avg_down_stride=True,
                 **kwargs):
        """Bottleneck block for ResNeSt."""
        super(Bottleneck, self).__init__(inplanes, planes, **kwargs)

        if groups == 1:
            width = self.planes
        else:
            width = math.floor(self.planes *
                               (base_width / base_channels)) * groups

        self.avg_down_stride = avg_down_stride and self.conv2_stride > 1

        self.norm1_name, norm1 = build_norm_layer(
            self.norm_cfg, width, postfix=1)
        self.norm3_name, norm3 = build_norm_layer(
            self.norm_cfg, self.planes * self.expansion, postfix=3)

        self.conv1 = build_conv_layer(
            self.conv_cfg,
            self.inplanes,
            width,
            kernel_size=1,
            stride=self.conv1_stride,
            bias=False)
        self.add_module(self.norm1_name, norm1)
        self.with_modulated_dcn = False
        self.conv2 = SplitAttentionConv2d(
            width,
            width,
            kernel_size=3,
            stride=1 if self.avg_down_stride else self.conv2_stride,
            padding=self.dilation,
            dilation=self.dilation,
            groups=groups,
            radix=radix,
            reduction_factor=reduction_factor,
            conv_cfg=self.conv_cfg,
            norm_cfg=self.norm_cfg,
            dcn=self.dcn)
        delattr(self, self.norm2_name)

        if self.avg_down_stride:
            self.avd_layer = nn.AvgPool2d(3, self.conv2_stride, padding=1)

        self.conv3 = build_conv_layer(
            self.conv_cfg,
            width,
            self.planes * self.expansion,
            kernel_size=1,
            bias=False)
        self.add_module(self.norm3_name, norm3)

    def forward(self, x):

        def _inner_forward(x):
            identity = x

            out = self.conv1(x)
            out = self.norm1(out)
            out = self.relu(out)

            if self.with_plugins:
                out = self.forward_plugin(out, self.after_conv1_plugin_names)

            out = self.conv2(out)

            if self.avg_down_stride:
                out = self.avd_layer(out)

            if self.with_plugins:
                out = self.forward_plugin(out, self.after_conv2_plugin_names)

            out = self.conv3(out)
            out = self.norm3(out)

            if self.with_plugins:
                out = self.forward_plugin(out, self.after_conv3_plugin_names)

            if self.downsample is not None:
                identity = self.downsample(x)

            out += identity

            return out

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

        out = self.relu(out)

        return out


@BACKBONES.register_module()
class ResNeSt(ResNetV1d):
    """ResNeSt backbone.

    Args:
        groups (int): Number of groups of Bottleneck. Default: 1
        base_width (int): Base width of Bottleneck. Default: 4
        radix (int): Radix of SpltAtConv2d. Default: 2
        reduction_factor (int): Reduction factor of inter_channels in
            SplitAttentionConv2d. Default: 4.
        avg_down_stride (bool): Whether to use average pool for stride in
            Bottleneck. Default: True.
        kwargs (dict): Keyword arguments for ResNet.
    """

    arch_settings = {
        50: (Bottleneck, (3, 4, 6, 3)),
        101: (Bottleneck, (3, 4, 23, 3)),
        152: (Bottleneck, (3, 8, 36, 3)),
        200: (Bottleneck, (3, 24, 36, 3))
    }

    def __init__(self,
                 groups=1,
                 base_width=4,
                 radix=2,
                 reduction_factor=4,
                 avg_down_stride=True,
                 **kwargs):
        self.groups = groups
        self.base_width = base_width
        self.radix = radix
        self.reduction_factor = reduction_factor
        self.avg_down_stride = avg_down_stride
        super(ResNeSt, self).__init__(**kwargs)

    def make_res_layer(self, **kwargs):
        """Pack all blocks in a stage into a ``ResLayer``."""
        return ResLayer(
            groups=self.groups,
            base_width=self.base_width,
            base_channels=self.base_channels,
            radix=self.radix,
            reduction_factor=self.reduction_factor,
            avg_down_stride=self.avg_down_stride,
            **kwargs)