File size: 10,040 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
'''
 * Copyright (c) 2023 Salesforce, Inc.
 * All rights reserved.
 * SPDX-License-Identifier: Apache License 2.0
 * For full license text, see LICENSE.txt file in the repo root or http://www.apache.org/licenses/
 * By Can Qin
 * Modified from ControlNet repo: https://github.com/lllyasviel/ControlNet
 * Copyright (c) 2023 Lvmin Zhang and Maneesh Agrawala
'''

import os
import sys
import torch
import torch.nn as nn
import torch.utils.model_zoo as model_zoo
from  torch.nn import  functional as F


class BlockTypeA(nn.Module):
    def __init__(self, in_c1, in_c2, out_c1, out_c2, upscale = True):
        super(BlockTypeA, self).__init__()
        self.conv1 = nn.Sequential(
            nn.Conv2d(in_c2, out_c2, kernel_size=1),
            nn.BatchNorm2d(out_c2),
            nn.ReLU(inplace=True)
        )
        self.conv2 = nn.Sequential(
            nn.Conv2d(in_c1, out_c1, kernel_size=1),
            nn.BatchNorm2d(out_c1),
            nn.ReLU(inplace=True)
        )
        self.upscale = upscale

    def forward(self, a, b):
        b = self.conv1(b)
        a = self.conv2(a)
        if self.upscale:
             b = F.interpolate(b, scale_factor=2.0, mode='bilinear', align_corners=True)
        return torch.cat((a, b), dim=1)


class BlockTypeB(nn.Module):
    def __init__(self, in_c, out_c):
        super(BlockTypeB, self).__init__()
        self.conv1 = nn.Sequential(
            nn.Conv2d(in_c, in_c,  kernel_size=3, padding=1),
            nn.BatchNorm2d(in_c),
            nn.ReLU()
        )
        self.conv2 = nn.Sequential(
            nn.Conv2d(in_c, out_c, kernel_size=3, padding=1),
            nn.BatchNorm2d(out_c),
            nn.ReLU()
        )

    def forward(self, x):
        x = self.conv1(x) + x
        x = self.conv2(x)
        return x

class BlockTypeC(nn.Module):
    def __init__(self, in_c, out_c):
        super(BlockTypeC, self).__init__()
        self.conv1 = nn.Sequential(
            nn.Conv2d(in_c, in_c,  kernel_size=3, padding=5, dilation=5),
            nn.BatchNorm2d(in_c),
            nn.ReLU()
        )
        self.conv2 = nn.Sequential(
            nn.Conv2d(in_c, in_c,  kernel_size=3, padding=1),
            nn.BatchNorm2d(in_c),
            nn.ReLU()
        )
        self.conv3 = nn.Conv2d(in_c, out_c, kernel_size=1)

    def forward(self, x):
        x = self.conv1(x)
        x = self.conv2(x)
        x = self.conv3(x)
        return x

def _make_divisible(v, divisor, min_value=None):
    """
    This function is taken from the original tf repo.
    It ensures that all layers have a channel number that is divisible by 8
    It can be seen here:
    https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet/mobilenet.py
    :param v:
    :param divisor:
    :param min_value:
    :return:
    """
    if min_value is None:
        min_value = divisor
    new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
    # Make sure that round down does not go down by more than 10%.
    if new_v < 0.9 * v:
        new_v += divisor
    return new_v


class ConvBNReLU(nn.Sequential):
    def __init__(self, in_planes, out_planes, kernel_size=3, stride=1, groups=1):
        self.channel_pad = out_planes - in_planes
        self.stride = stride
        #padding = (kernel_size - 1) // 2

        # TFLite uses slightly different padding than PyTorch
        if stride == 2:
            padding = 0
        else:
            padding = (kernel_size - 1) // 2

        super(ConvBNReLU, self).__init__(
            nn.Conv2d(in_planes, out_planes, kernel_size, stride, padding, groups=groups, bias=False),
            nn.BatchNorm2d(out_planes),
            nn.ReLU6(inplace=True)
        )
        self.max_pool = nn.MaxPool2d(kernel_size=stride, stride=stride)


    def forward(self, x):
        # TFLite uses  different padding
        if self.stride == 2:
            x = F.pad(x, (0, 1, 0, 1), "constant", 0)
            #print(x.shape)

        for module in self:
            if not isinstance(module, nn.MaxPool2d):
                x = module(x)
        return x


class InvertedResidual(nn.Module):
    def __init__(self, inp, oup, stride, expand_ratio):
        super(InvertedResidual, self).__init__()
        self.stride = stride
        assert stride in [1, 2]

        hidden_dim = int(round(inp * expand_ratio))
        self.use_res_connect = self.stride == 1 and inp == oup

        layers = []
        if expand_ratio != 1:
            # pw
            layers.append(ConvBNReLU(inp, hidden_dim, kernel_size=1))
        layers.extend([
            # dw
            ConvBNReLU(hidden_dim, hidden_dim, stride=stride, groups=hidden_dim),
            # pw-linear
            nn.Conv2d(hidden_dim, oup, 1, 1, 0, bias=False),
            nn.BatchNorm2d(oup),
        ])
        self.conv = nn.Sequential(*layers)

    def forward(self, x):
        if self.use_res_connect:
            return x + self.conv(x)
        else:
            return self.conv(x)


class MobileNetV2(nn.Module):
    def __init__(self, pretrained=True):
        """
        MobileNet V2 main class
        Args:
            num_classes (int): Number of classes
            width_mult (float): Width multiplier - adjusts number of channels in each layer by this amount
            inverted_residual_setting: Network structure
            round_nearest (int): Round the number of channels in each layer to be a multiple of this number
            Set to 1 to turn off rounding
            block: Module specifying inverted residual building block for mobilenet
        """
        super(MobileNetV2, self).__init__()

        block = InvertedResidual
        input_channel = 32
        last_channel = 1280
        width_mult = 1.0
        round_nearest = 8

        inverted_residual_setting = [
            # t, c, n, s
            [1, 16, 1, 1],
            [6, 24, 2, 2],
            [6, 32, 3, 2],
            [6, 64, 4, 2],
            [6, 96, 3, 1],
            #[6, 160, 3, 2],
            #[6, 320, 1, 1],
        ]

        # only check the first element, assuming user knows t,c,n,s are required
        if len(inverted_residual_setting) == 0 or len(inverted_residual_setting[0]) != 4:
            raise ValueError("inverted_residual_setting should be non-empty "
                             "or a 4-element list, got {}".format(inverted_residual_setting))

        # building first layer
        input_channel = _make_divisible(input_channel * width_mult, round_nearest)
        self.last_channel = _make_divisible(last_channel * max(1.0, width_mult), round_nearest)
        features = [ConvBNReLU(4, input_channel, stride=2)]
        # building inverted residual blocks
        for t, c, n, s in inverted_residual_setting:
            output_channel = _make_divisible(c * width_mult, round_nearest)
            for i in range(n):
                stride = s if i == 0 else 1
                features.append(block(input_channel, output_channel, stride, expand_ratio=t))
                input_channel = output_channel

        self.features = nn.Sequential(*features)
        self.fpn_selected = [1, 3, 6, 10, 13]
        # weight initialization
        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight, mode='fan_out')
                if m.bias is not None:
                    nn.init.zeros_(m.bias)
            elif isinstance(m, nn.BatchNorm2d):
                nn.init.ones_(m.weight)
                nn.init.zeros_(m.bias)
            elif isinstance(m, nn.Linear):
                nn.init.normal_(m.weight, 0, 0.01)
                nn.init.zeros_(m.bias)
        if pretrained:
           self._load_pretrained_model()

    def _forward_impl(self, x):
        # This exists since TorchScript doesn't support inheritance, so the superclass method
        # (this one) needs to have a name other than `forward` that can be accessed in a subclass
        fpn_features = []
        for i, f in enumerate(self.features):
            if i > self.fpn_selected[-1]:
                break
            x = f(x)
            if i in self.fpn_selected:
                fpn_features.append(x)

        c1, c2, c3, c4, c5 = fpn_features
        return c1, c2, c3, c4, c5


    def forward(self, x):
        return self._forward_impl(x)

    def _load_pretrained_model(self):
        pretrain_dict = model_zoo.load_url('https://download.pytorch.org/models/mobilenet_v2-b0353104.pth')
        model_dict = {}
        state_dict = self.state_dict()
        for k, v in pretrain_dict.items():
            if k in state_dict:
                model_dict[k] = v
        state_dict.update(model_dict)
        self.load_state_dict(state_dict)


class MobileV2_MLSD_Large(nn.Module):
    def __init__(self):
        super(MobileV2_MLSD_Large, self).__init__()

        self.backbone = MobileNetV2(pretrained=False)
        ## A, B
        self.block15 = BlockTypeA(in_c1= 64, in_c2= 96,
                                  out_c1= 64, out_c2=64,
                                  upscale=False)
        self.block16 = BlockTypeB(128, 64)

        ## A, B
        self.block17 = BlockTypeA(in_c1 = 32,  in_c2 = 64,
                                  out_c1= 64,  out_c2= 64)
        self.block18 = BlockTypeB(128, 64)

        ## A, B
        self.block19 = BlockTypeA(in_c1=24, in_c2=64,
                                  out_c1=64, out_c2=64)
        self.block20 = BlockTypeB(128, 64)

        ## A, B, C
        self.block21 = BlockTypeA(in_c1=16, in_c2=64,
                                  out_c1=64, out_c2=64)
        self.block22 = BlockTypeB(128, 64)

        self.block23 = BlockTypeC(64, 16)

    def forward(self, x):
        c1, c2, c3, c4, c5 = self.backbone(x)

        x = self.block15(c4, c5)
        x = self.block16(x)

        x = self.block17(c3, x)
        x = self.block18(x)

        x = self.block19(c2, x)
        x = self.block20(x)

        x = self.block21(c1, x)
        x = self.block22(x)
        x = self.block23(x)
        x = x[:, 7:, :, :]

        return x