Skip to content

deeplabcut.pose_estimation_pytorch.models.modules.conv_block

The code is based on DEKR: https://github.com/HRNet/DEKR/tree/main

Classes:

Name Description
AdaptBlock

Adaptive Residual Block with Deformable Convolution.
BaseBlock

Abstract Base class for defining custom blocks.
BasicBlock

Basic Residual Block.
Bottleneck

Bottleneck Residual Block.

AdaptBlock

Bases: BaseBlock

Adaptive Residual Block with Deformable Convolution.

This class defines an adaptive residual block with deformable convolution used in HigherHRNet.

Attributes:

Name Type Description
expansion int

The expansion factor used in the block.

Parameters:

Name Type Description Default

in_channels

int

Number of input channels.

 required

out_channels

int

Number of output channels.

 required

stride

int

Stride value for the convolutional layers. Default is 1.

 1

downsample

Module | None

Downsample layer to be used in the residual connection. Default is None.

 None

dilation

int

Dilation rate for the convolutional layers. Default is 1.

 1

deformable_groups

int

Number of deformable groups in the deformable convolution. Default is 1.

 1

Methods:

Name Description
forward

Forward pass through the AdaptBlock.
Source code in deeplabcut/pose_estimation_pytorch/models/modules/conv_block.py 
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
@BLOCKS.register_module
class AdaptBlock(BaseBlock):
    """Adaptive Residual Block with Deformable Convolution.

    This class defines an adaptive residual block with deformable convolution used in HigherHRNet.

    Attributes:
        expansion: The expansion factor used in the block.

    Args:
        in_channels: Number of input channels.
        out_channels: Number of output channels.
        stride: Stride value for the convolutional layers. Default is 1.
        downsample: Downsample layer to be used in the residual connection. Default is None.
        dilation: Dilation rate for the convolutional layers. Default is 1.
        deformable_groups: Number of deformable groups in the deformable convolution. Default is 1.
    """

    expansion: int = 1

    def __init__(
        self,
        in_channels: int,
        out_channels: int,
        stride: int = 1,
        downsample: nn.Module | None = None,
        dilation: int = 1,
        deformable_groups: int = 1,
    ):
        super().__init__()
        regular_matrix = torch.tensor([[-1, -1, -1, 0, 0, 0, 1, 1, 1], [-1, 0, 1, -1, 0, 1, -1, 0, 1]])
        self.register_buffer("regular_matrix", regular_matrix.float())
        self.downsample = downsample
        self.transform_matrix_conv = nn.Conv2d(in_channels, 4, 3, 1, 1, bias=True)
        self.translation_conv = nn.Conv2d(in_channels, 2, 3, 1, 1, bias=True)
        self.adapt_conv = ops.DeformConv2d(
            in_channels,
            out_channels,
            kernel_size=3,
            stride=stride,
            padding=dilation,
            dilation=dilation,
            bias=False,
            groups=deformable_groups,
        )
        self.bn = nn.BatchNorm2d(out_channels, momentum=self.bn_momentum)
        self.relu = nn.ReLU(inplace=True)

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        """Forward pass through the AdaptBlock.

        Args:
            x: Input tensor.

        Returns:
            Output tensor.
        """
        residual = x

        N, _, H, W = x.shape
        transform_matrix = self.transform_matrix_conv(x)
        transform_matrix = transform_matrix.permute(0, 2, 3, 1).reshape((N * H * W, 2, 2))
        offset = torch.matmul(transform_matrix, self.regular_matrix)
        offset = offset - self.regular_matrix
        offset = offset.transpose(1, 2).reshape((N, H, W, 18)).permute(0, 3, 1, 2)

        translation = self.translation_conv(x)
        offset[:, 0::2, :, :] += translation[:, 0:1, :, :]
        offset[:, 1::2, :, :] += translation[:, 1:2, :, :]

        out = self.adapt_conv(x, offset)
        out = self.bn(out)

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

        out += residual
        out = self.relu(out)

        return out

forward 

forward(x: Tensor) -> torch.Tensor

Forward pass through the AdaptBlock.

Parameters:

Name Type Description Default

x

Tensor

Input tensor.

 required

Returns:

Type Description
Tensor

Output tensor.
Source code in deeplabcut/pose_estimation_pytorch/models/modules/conv_block.py 
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
def forward(self, x: torch.Tensor) -> torch.Tensor:
    """Forward pass through the AdaptBlock.

    Args:
        x: Input tensor.

    Returns:
        Output tensor.
    """
    residual = x

    N, _, H, W = x.shape
    transform_matrix = self.transform_matrix_conv(x)
    transform_matrix = transform_matrix.permute(0, 2, 3, 1).reshape((N * H * W, 2, 2))
    offset = torch.matmul(transform_matrix, self.regular_matrix)
    offset = offset - self.regular_matrix
    offset = offset.transpose(1, 2).reshape((N, H, W, 18)).permute(0, 3, 1, 2)

    translation = self.translation_conv(x)
    offset[:, 0::2, :, :] += translation[:, 0:1, :, :]
    offset[:, 1::2, :, :] += translation[:, 1:2, :, :]

    out = self.adapt_conv(x, offset)
    out = self.bn(out)

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

    out += residual
    out = self.relu(out)

    return out

BaseBlock

Bases: ABC, Module

Abstract Base class for defining custom blocks.

This class defines an abstract base class for creating custom blocks used in the HigherHRNet for Human Pose Estimation.

Attributes:

Name Type Description
bn_momentum

Batch normalization momentum.

Methods:

Name Description
forward

Abstract method for defining the forward pass of the block.
Source code in deeplabcut/pose_estimation_pytorch/models/modules/conv_block.py 
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
class BaseBlock(ABC, nn.Module):
    """Abstract Base class for defining custom blocks.

    This class defines an abstract base class for creating custom blocks used in the HigherHRNet for Human Pose
    Estimation.

    Attributes:
        bn_momentum: Batch normalization momentum.

    Methods:
        forward(x): Abstract method for defining the forward pass of the block.
    """

    def __init__(self):
        super().__init__()
        self.bn_momentum = 0.1

    @abstractmethod
    def forward(self, x: torch.Tensor):
        """Abstract method for defining the forward pass of the block.

        Args:
            x: Input tensor.

        Returns:
            Output tensor.
        """
        pass

    def _init_weights(self, pretrained: str | None):
        """Method for initializing block weights from pretrained models.

        Args:
            pretrained: Path to pretrained model weights.
        """
        if pretrained:
            self.load_state_dict(torch.load(pretrained))

forward abstractmethod 

forward(x: Tensor)

Abstract method for defining the forward pass of the block.

Parameters:

Name Type Description Default

x

Tensor

Input tensor.

 required

Returns:

Type Description

Output tensor.
Source code in deeplabcut/pose_estimation_pytorch/models/modules/conv_block.py 
43
44
45
46
47
48
49
50
51
52
53
@abstractmethod
def forward(self, x: torch.Tensor):
    """Abstract method for defining the forward pass of the block.

    Args:
        x: Input tensor.

    Returns:
        Output tensor.
    """
    pass

BasicBlock

Bases: BaseBlock

Basic Residual Block.

This class defines a basic residual block used in HigherHRNet.

Attributes:

Name Type Description
expansion int

The expansion factor used in the block.

Parameters:

Name Type Description Default

in_channels

int

Number of input channels.

 required

out_channels

int

Number of output channels.

 required

stride

int

Stride value for the convolutional layers. Default is 1.

 1

downsample

Module | None

Downsample layer to be used in the residual connection. Default is None.

 None

dilation

int

Dilation rate for the convolutional layers. Default is 1.

 1

Methods:

Name Description
forward

Forward pass through the BasicBlock.
Source code in deeplabcut/pose_estimation_pytorch/models/modules/conv_block.py 
 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
@BLOCKS.register_module
class BasicBlock(BaseBlock):
    """Basic Residual Block.

    This class defines a basic residual block used in HigherHRNet.

    Attributes:
        expansion: The expansion factor used in the block.

    Args:
        in_channels: Number of input channels.
        out_channels: Number of output channels.
        stride: Stride value for the convolutional layers. Default is 1.
        downsample: Downsample layer to be used in the residual connection. Default is None.
        dilation: Dilation rate for the convolutional layers. Default is 1.
    """

    expansion: int = 1

    def __init__(
        self,
        in_channels: int,
        out_channels: int,
        stride: int = 1,
        downsample: nn.Module | None = None,
        dilation: int = 1,
    ):
        super().__init__()
        self.conv1 = nn.Conv2d(
            in_channels,
            out_channels,
            kernel_size=3,
            stride=stride,
            padding=dilation,
            bias=False,
            dilation=dilation,
        )
        self.bn1 = nn.BatchNorm2d(out_channels, momentum=self.bn_momentum)
        self.relu = nn.ReLU(inplace=True)
        self.conv2 = nn.Conv2d(
            in_channels,
            out_channels,
            kernel_size=3,
            stride=stride,
            padding=dilation,
            bias=False,
            dilation=dilation,
        )
        self.bn2 = nn.BatchNorm2d(out_channels, momentum=self.bn_momentum)
        self.downsample = downsample
        self.stride = stride

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        """Forward pass through the BasicBlock.

        Args:
            x: Input tensor.

        Returns:
            Output tensor.
        """
        residual = x

        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)

        out = self.conv2(out)
        out = self.bn2(out)

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

        out += residual
        out = self.relu(out)

        return out

forward 

forward(x: Tensor) -> torch.Tensor

Forward pass through the BasicBlock.

Parameters:

Name Type Description Default

x

Tensor

Input tensor.

 required

Returns:

Type Description
Tensor

Output tensor.
Source code in deeplabcut/pose_estimation_pytorch/models/modules/conv_block.py 
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
def forward(self, x: torch.Tensor) -> torch.Tensor:
    """Forward pass through the BasicBlock.

    Args:
        x: Input tensor.

    Returns:
        Output tensor.
    """
    residual = x

    out = self.conv1(x)
    out = self.bn1(out)
    out = self.relu(out)

    out = self.conv2(out)
    out = self.bn2(out)

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

    out += residual
    out = self.relu(out)

    return out

Bottleneck

Bases: BaseBlock

Bottleneck Residual Block.

This class defines a bottleneck residual block used in HigherHRNet.

Attributes:

Name Type Description
expansion int

The expansion factor used in the block.

Parameters:

Name Type Description Default

in_channels

int

Number of input channels.

 required

out_channels

int

Number of output channels.

 required

stride

int

Stride value for the convolutional layers. Default is 1.

 1

downsample

Module | None

Downsample layer to be used in the residual connection. Default is None.

 None

dilation

int

Dilation rate for the convolutional layers. Default is 1.

 1

Methods:

Name Description
forward

Forward pass through the Bottleneck block.
Source code in deeplabcut/pose_estimation_pytorch/models/modules/conv_block.py 
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
@BLOCKS.register_module
class Bottleneck(BaseBlock):
    """Bottleneck Residual Block.

    This class defines a bottleneck residual block used in HigherHRNet.

    Attributes:
        expansion: The expansion factor used in the block.

    Args:
        in_channels: Number of input channels.
        out_channels: Number of output channels.
        stride: Stride value for the convolutional layers. Default is 1.
        downsample: Downsample layer to be used in the residual connection. Default is None.
        dilation: Dilation rate for the convolutional layers. Default is 1.
    """

    expansion: int = 4

    def __init__(
        self,
        in_channels: int,
        out_channels: int,
        stride: int = 1,
        downsample: nn.Module | None = None,
        dilation: int = 1,
    ):
        super().__init__()
        self.conv1 = nn.Conv2d(in_channels, out_channels, kernel_size=1, bias=False)
        self.bn1 = nn.BatchNorm2d(out_channels, momentum=self.bn_momentum)
        self.conv2 = nn.Conv2d(
            out_channels,
            out_channels,
            kernel_size=3,
            stride=stride,
            padding=dilation,
            bias=False,
            dilation=dilation,
        )
        self.bn2 = nn.BatchNorm2d(out_channels, momentum=self.bn_momentum)
        self.conv3 = nn.Conv2d(out_channels, out_channels * self.expansion, kernel_size=1, bias=False)
        self.bn3 = nn.BatchNorm2d(out_channels * self.expansion, momentum=self.bn_momentum)
        self.relu = nn.ReLU(inplace=True)
        self.downsample = downsample
        self.stride = stride

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        """Forward pass through the Bottleneck block.

        Args:
            x : Input tensor.

        Returns:
            Output tensor.
        """
        residual = x

        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)

        out = self.conv2(out)
        out = self.bn2(out)
        out = self.relu(out)

        out = self.conv3(out)
        out = self.bn3(out)

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

        out += residual
        out = self.relu(out)

        return out

forward 

forward(x: Tensor) -> torch.Tensor

Forward pass through the Bottleneck block.

Parameters:

Name Type Description Default

x

Input tensor.

 required

Returns:

Type Description
Tensor

Output tensor.
Source code in deeplabcut/pose_estimation_pytorch/models/modules/conv_block.py 
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
def forward(self, x: torch.Tensor) -> torch.Tensor:
    """Forward pass through the Bottleneck block.

    Args:
        x : Input tensor.

    Returns:
        Output tensor.
    """
    residual = x

    out = self.conv1(x)
    out = self.bn1(out)
    out = self.relu(out)

    out = self.conv2(out)
    out = self.bn2(out)
    out = self.relu(out)

    out = self.conv3(out)
    out = self.bn3(out)

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

    out += residual
    out = self.relu(out)

    return out