File size: 10,007 Bytes
231edce
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import torch
import torch.nn as nn

from torch import Tensor
from torchvision.models.video.resnet import BasicBlock, Bottleneck, Conv3DSimple, Conv3DNoTemporal, Conv2Plus1D
from typing import Callable, List, Sequence, Type, Union


# TODO: upload models and load them
model_urls = {
    "r2plus1d_34_8_ig65m": "https://github.com/moabitcoin/ig65m-pytorch/releases/download/v1.0.0/r2plus1d_34_clip8_ig65m_from_scratch-9bae36ae.pth",  # noqa: E501
    "r2plus1d_34_32_ig65m": "https://github.com/moabitcoin/ig65m-pytorch/releases/download/v1.0.0/r2plus1d_34_clip32_ig65m_from_scratch-449a7af9.pth",  # noqa: E501
    "r2plus1d_34_8_kinetics": "https://github.com/moabitcoin/ig65m-pytorch/releases/download/v1.0.0/r2plus1d_34_clip8_ft_kinetics_from_ig65m-0aa0550b.pth",  # noqa: E501
    "r2plus1d_34_32_kinetics": "https://github.com/moabitcoin/ig65m-pytorch/releases/download/v1.0.0/r2plus1d_34_clip32_ft_kinetics_from_ig65m-ade133f1.pth",  # noqa: E501
    "r2plus1d_152_ig65m_32frms": "https://github.com/bjuncek/VMZ/releases/download/test_models/r2plus1d_152_ig65m_from_scratch_f106380637.pth",
    "r2plus1d_152_ig_ft_kinetics_32frms": "https://github.com/bjuncek/VMZ/releases/download/test_models/r2plus1d_152_ft_kinetics_from_ig65m_f107107466.pth",
    "r2plus1d_152_sports1m_32frms": "",
    "r2plus1d_152_sports1m_ft_kinetics_32frms": "https://github.com/bjuncek/VMZ/releases/download/test_models/r2plus1d_152_ft_kinetics_from_sports1m_f128957437.pth",
    "ir_csn_152_ig65m_32frms": "https://github.com/bjuncek/VMZ/releases/download/test_models/irCSN_152_ig65m_from_scratch_f125286141.pth",
    "ir_csn_152_ig_ft_kinetics_32frms": "https://github.com/bjuncek/VMZ/releases/download/test_models/irCSN_152_ft_kinetics_from_ig65m_f126851907.pth",
    "ir_csn_152_sports1m_32frms": "https://github.com/bjuncek/VMZ/releases/download/test_models/irCSN_152_Sports1M_from_scratch_f99918785.pth",
    "ir_csn_152_sports1m_ft_kinetics_32frms": "https://github.com/bjuncek/VMZ/releases/download/test_models/irCSN_152_ft_kinetics_from_Sports1M_f101599884.pth",
    "ip_csn_152_ig65m_32frms": "https://github.com/bjuncek/VMZ/releases/download/test_models/ipCSN_152_ig65m_from_scratch_f130601052.pth",
    "ip_csn_152_ig_ft_kinetics_32frms": "https://github.com/bjuncek/VMZ/releases/download/test_models/ipCSN_152_ft_kinetics_from_ig65m_f133090949.pth",
    "ip_csn_152_sports1m_32frms": "https://github.com/bjuncek/VMZ/releases/download/test_models/ipCSN_152_Sports1M_from_scratch_f111018543.pth",
    "ip_csn_152_sports1m_ft_kinetics_32frms": "https://github.com/bjuncek/VMZ/releases/download/test_models/ipCSN_152_ft_kinetics_from_Sports1M_f111279053.pth",
}


class VideoResNet(nn.Module):
    def __init__(
        self,
        block: Type[Union[BasicBlock, Bottleneck]],
        conv_makers: Sequence[Type[Union[Conv3DSimple, Conv3DNoTemporal, Conv2Plus1D]]],
        layers: List[int],
        stem: Callable[..., nn.Module],
        num_classes: int = 400,
        zero_init_residual: bool = False,
    ) -> None:
        """Generic resnet video generator.

        Args:
            block (Type[Union[BasicBlock, Bottleneck]]): resnet building block
            conv_makers (List[Type[Union[Conv3DSimple, Conv3DNoTemporal, Conv2Plus1D]]]): generator
                function for each layer
            layers (List[int]): number of blocks per layer
            stem (Callable[..., nn.Module]): module specifying the ResNet stem.
            num_classes (int, optional): Dimension of the final FC layer. Defaults to 400.
            zero_init_residual (bool, optional): Zero init bottleneck residual BN. Defaults to False.
        """
        super().__init__()
        self.inplanes = 64

        self.stem = stem()

        self.layer1 = self._make_layer(block, conv_makers[0], 64, layers[0], stride=1)
        self.layer2 = self._make_layer(block, conv_makers[1], 128, layers[1], stride=2)
        self.layer3 = self._make_layer(block, conv_makers[2], 256, layers[2], stride=2)
        self.layer4 = self._make_layer(block, conv_makers[3], 512, layers[3], stride=2)

        self.avgpool = nn.AdaptiveAvgPool3d((1, 1, 1))
        self.fc = nn.Linear(512 * block.expansion, num_classes)

        # init weights
        for m in self.modules():
            if isinstance(m, nn.Conv3d):
                nn.init.kaiming_normal_(m.weight, mode="fan_out", nonlinearity="relu")
                if m.bias is not None:
                    nn.init.constant_(m.bias, 0)
            elif isinstance(m, nn.BatchNorm3d):
                nn.init.constant_(m.weight, 1)
                nn.init.constant_(m.bias, 0)
            elif isinstance(m, nn.Linear):
                nn.init.normal_(m.weight, 0, 0.01)
                nn.init.constant_(m.bias, 0)

        if zero_init_residual:
            for m in self.modules():
                if isinstance(m, Bottleneck):
                    nn.init.constant_(m.bn3.weight, 0)  # type: ignore[union-attr, arg-type]

    def forward(self, x: Tensor) -> Tensor:
        x = self.stem(x)

        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.layer4(x)

        x = self.avgpool(x)
        x = self.fc(x)

        return x

    def _make_layer(
        self,
        block: Type[Union[BasicBlock, Bottleneck]],
        conv_builder: Type[Union[Conv3DSimple, Conv3DNoTemporal, Conv2Plus1D]],
        planes: int,
        blocks: int,
        stride: int = 1,
    ) -> nn.Sequential:
        downsample = None

        if stride != 1 or self.inplanes != planes * block.expansion:
            ds_stride = conv_builder.get_downsample_stride(stride)
            downsample = nn.Sequential(
                nn.Conv3d(self.inplanes, planes * block.expansion, kernel_size=1, stride=ds_stride, bias=False),
                nn.BatchNorm3d(planes * block.expansion),
            )
        layers = []
        layers.append(block(self.inplanes, planes, conv_builder, stride, downsample))

        self.inplanes = planes * block.expansion
        for i in range(1, blocks):
            layers.append(block(self.inplanes, planes, conv_builder))

        return nn.Sequential(*layers)


def _generic_resnet(arch, pretrained=False, progress=False, **kwargs):
    model = VideoResNet(**kwargs)

    # We need exact Caffe2 momentum for BatchNorm scaling
    for m in model.modules():
        if isinstance(m, nn.BatchNorm3d):
            m.eps = 1e-3
            m.momentum = 0.9

    if pretrained:
        state_dict = torch.hub.load_state_dict_from_url(
            model_urls[arch], progress=progress
        )
        model.load_state_dict(state_dict)

    return model


class BasicStem_Pool(nn.Sequential):
    def __init__(self):
        super(BasicStem_Pool, self).__init__(
            nn.Conv3d(
                3,
                64,
                kernel_size=(3, 7, 7),
                stride=(1, 2, 2),
                padding=(1, 3, 3),
                bias=False,
            ),
            nn.BatchNorm3d(64),
            nn.ReLU(inplace=True),
            nn.MaxPool3d(kernel_size=(1, 3, 3), stride=(1, 2, 2), padding=(0, 1, 1)),
        )


class R2Plus1dStem_Pool(nn.Sequential):
    """R(2+1)D stem is different than the default one as it uses separated 3D convolution
    """

    def __init__(self):
        super(R2Plus1dStem_Pool, self).__init__(
            nn.Conv3d(
                3,
                45,
                kernel_size=(1, 7, 7),
                stride=(1, 2, 2),
                padding=(0, 3, 3),
                bias=False,
            ),
            nn.BatchNorm3d(45),
            nn.ReLU(inplace=True),
            nn.Conv3d(
                45,
                64,
                kernel_size=(3, 1, 1),
                stride=(1, 1, 1),
                padding=(1, 0, 0),
                bias=False,
            ),
            nn.BatchNorm3d(64),
            nn.ReLU(inplace=True),
            nn.MaxPool3d(kernel_size=(1, 3, 3), stride=(1, 2, 2), padding=(0, 1, 1)),
        )


class Conv3DDepthwise(nn.Conv3d):
    def __init__(self, in_planes, out_planes, midplanes=None, stride=1, padding=1):

        assert in_planes == out_planes
        super(Conv3DDepthwise, self).__init__(
            in_channels=in_planes,
            out_channels=out_planes,
            kernel_size=(3, 3, 3),
            stride=stride,
            padding=padding,
            groups=in_planes,
            bias=False,
        )

    @staticmethod
    def get_downsample_stride(stride):
        return (stride, stride, stride)


class IPConv3DDepthwise(nn.Sequential):
    def __init__(self, in_planes, out_planes, midplanes, stride=1, padding=1):

        assert in_planes == out_planes
        super(IPConv3DDepthwise, self).__init__(
            nn.Conv3d(in_planes, out_planes, kernel_size=1, bias=False),
            nn.BatchNorm3d(out_planes),
            # nn.ReLU(inplace=True),
            Conv3DDepthwise(out_planes, out_planes, None, stride),
        )

    @staticmethod
    def get_downsample_stride(stride):
        return (stride, stride, stride)


class Conv2Plus1D(nn.Sequential):
    def __init__(self, in_planes, out_planes, midplanes, stride=1, padding=1):

        midplanes = (in_planes * out_planes * 3 * 3 * 3) // (
            in_planes * 3 * 3 + 3 * out_planes
        )
        super(Conv2Plus1D, self).__init__(
            nn.Conv3d(
                in_planes,
                midplanes,
                kernel_size=(1, 3, 3),
                stride=(1, stride, stride),
                padding=(0, padding, padding),
                bias=False,
            ),
            nn.BatchNorm3d(midplanes),
            nn.ReLU(inplace=True),
            nn.Conv3d(
                midplanes,
                out_planes,
                kernel_size=(3, 1, 1),
                stride=(stride, 1, 1),
                padding=(padding, 0, 0),
                bias=False,
            ),
        )

    @staticmethod
    def get_downsample_stride(stride):
        return (stride, stride, stride)