modules.conv.DecoderNet2d¶
net = mdnc.modules.conv.DecoderNet2d(
channel, layers, out_size,
kernel_size=3, in_length=2, out_planes=1
)
This moule is a built-in model for 2D convolutional decoder network. This network could be used as a part of the auto-encoder, or just a network for up-sampling (or generating) data.
The network would up-sample the input data according to the network depth. The depth is given by the length of the argument layers. The network structure is shown in the following chart:
flowchart TB
u1["Block 1<br>Stack of layers[0] layers"]
u2["Block 2<br>Stack of layers[1] layers"]
ui["Block ...<br>Stack of ... layers"]
un["Block n<br>Stack of layers[n-1] layers"]
optional:::blockoptional
subgraph optional [Optional]
cin["Conv2d<br>with unsqueeze"]
end
u1 -->|up<br>sampling| u2 -->|up<br>sampling| ui -->|up<br>sampling| un
cin -.-> u1
linkStyle 0,1,2 stroke-width:4px, stroke:#080 ;
classDef blockoptional fill:none, stroke-dasharray:10,10, stroke:#9370DB, width:100;The argument layers is a sequence of int. For each block \(i\), it contains layers[i-1] repeated modern convolutional layers (see mdnc.modules.conv.ConvModern2d). Each up-sampling is configured by stride=2. The channel number would be doubled in the up-sampling route. An optional unsqueezer and convolutional layer could be prepended to the first layer when the argument in_length != None. This optional layer is used for converting the vector features in initial feature maps.
Arguments¶
Requries
| Argument | Type | Description |
|---|---|---|
channel | int | The channel number of the first hidden block (layer). After each down-sampling, the channel number would be doubled. |
layers | (int,) | A sequence of layer numbers for each block. Each number represents the number of convolutional layers of a stage (block). The stage numer, i.e. the depth of the network is the length of this list. |
out_size | int or(int, int) | The size of the output data. This argument needs to be specified by users, because the network needs to configure its layers according to the output size. |
kernel_size | int or(int, int) | The kernel size of each convolutional layer. |
in_length | int | The length of the input vector, if not set, the input needs to be feature maps. See the property input_size to check the input data size in this case. |
out_planes | int | The channel number of the output data. |
Operators¶
__call__¶
y = net(x)
The forward operator implemented by the forward() method. The input data is a tensor with a size determined by configurations. The output is a 2D tensor. The channel number of the output is specified by the argument out_planes.
Requries
| Argument | Type | Description |
|---|---|---|
x | torch.Tensor | A tensor,
|
Returns
| Argument | Description |
|---|---|
y | A 2D tensor, the size should be (B, C, L1, L2), where B is the batch size, C is the input channel number, and (L1, L2) is the output data size specified by the argument out_size. |
Properties¶
nlayers¶
net.nlayers
The total number of convolutional layers along the depth of the network. This value would not take the fully-connected layer into consideration.
input_size¶
net.input_size
The size of the input data size (a tuple). This property is useful when in_length is None. In this case, the input size is determined by the network.
Warning
This size contains the channel number (as the first element), because the input channel number is also determined by network when in_length is None.
Examples¶
Example 1
1 2 3 4 5 6 | |
The number of convolutional layers along the depth is 18.
The input size is (32,).
----------------------------------------------------------------
Layer (type) Output Shape Param #
================================================================
Conv2d-1 [-1, 1024, 2, 2] 132,096
Conv2d-2 [-1, 1024, 2, 2] 9,437,184
InstanceNorm2d-3 [-1, 1024, 2, 2] 2,048
PReLU-4 [-1, 1024, 2, 2] 1,024
Conv2d-5 [-1, 512, 2, 2] 4,718,592
_ConvModernNd-6 [-1, 512, 2, 2] 0
InstanceNorm2d-7 [-1, 512, 2, 2] 1,024
PReLU-8 [-1, 512, 2, 2] 512
Conv2d-9 [-1, 512, 2, 2] 2,359,296
_ConvModernNd-10 [-1, 512, 2, 2] 0
InstanceNorm2d-11 [-1, 512, 2, 2] 1,024
PReLU-12 [-1, 512, 2, 2] 512
Upsample-13 [-1, 512, 4, 4] 0
Conv2d-14 [-1, 512, 4, 4] 2,359,296
_ConvModernNd-15 [-1, 512, 4, 4] 0
_BlockConvStkNd-16 [-1, 512, 4, 4] 0
InstanceNorm2d-17 [-1, 512, 4, 4] 1,024
PReLU-18 [-1, 512, 4, 4] 512
Conv2d-19 [-1, 256, 4, 4] 1,179,648
_ConvModernNd-20 [-1, 256, 4, 4] 0
InstanceNorm2d-21 [-1, 256, 4, 4] 512
PReLU-22 [-1, 256, 4, 4] 256
Conv2d-23 [-1, 256, 4, 4] 589,824
_ConvModernNd-24 [-1, 256, 4, 4] 0
InstanceNorm2d-25 [-1, 256, 4, 4] 512
PReLU-26 [-1, 256, 4, 4] 256
Upsample-27 [-1, 256, 8, 8] 0
Conv2d-28 [-1, 256, 8, 8] 589,824
_ConvModernNd-29 [-1, 256, 8, 8] 0
_BlockConvStkNd-30 [-1, 256, 8, 8] 0
InstanceNorm2d-31 [-1, 256, 8, 8] 512
PReLU-32 [-1, 256, 8, 8] 256
Conv2d-33 [-1, 128, 8, 8] 294,912
_ConvModernNd-34 [-1, 128, 8, 8] 0
InstanceNorm2d-35 [-1, 128, 8, 8] 256
PReLU-36 [-1, 128, 8, 8] 128
Conv2d-37 [-1, 128, 8, 8] 147,456
_ConvModernNd-38 [-1, 128, 8, 8] 0
InstanceNorm2d-39 [-1, 128, 8, 8] 256
PReLU-40 [-1, 128, 8, 8] 128
Upsample-41 [-1, 128, 16, 16] 0
Conv2d-42 [-1, 128, 16, 16] 147,456
_ConvModernNd-43 [-1, 128, 16, 16] 0
_BlockConvStkNd-44 [-1, 128, 16, 16] 0
InstanceNorm2d-45 [-1, 128, 16, 16] 256
PReLU-46 [-1, 128, 16, 16] 128
Conv2d-47 [-1, 64, 16, 16] 73,728
_ConvModernNd-48 [-1, 64, 16, 16] 0
InstanceNorm2d-49 [-1, 64, 16, 16] 128
PReLU-50 [-1, 64, 16, 16] 64
Conv2d-51 [-1, 64, 16, 16] 36,864
_ConvModernNd-52 [-1, 64, 16, 16] 0
InstanceNorm2d-53 [-1, 64, 16, 16] 128
PReLU-54 [-1, 64, 16, 16] 64
Upsample-55 [-1, 64, 32, 32] 0
Conv2d-56 [-1, 64, 32, 32] 36,864
_ConvModernNd-57 [-1, 64, 32, 32] 0
_BlockConvStkNd-58 [-1, 64, 32, 32] 0
InstanceNorm2d-59 [-1, 64, 32, 32] 128
PReLU-60 [-1, 64, 32, 32] 64
Conv2d-61 [-1, 64, 32, 32] 36,864
_ConvModernNd-62 [-1, 64, 32, 32] 0
InstanceNorm2d-63 [-1, 64, 32, 32] 128
PReLU-64 [-1, 64, 32, 32] 64
Conv2d-65 [-1, 64, 32, 32] 36,864
_ConvModernNd-66 [-1, 64, 32, 32] 0
InstanceNorm2d-67 [-1, 64, 32, 32] 128
PReLU-68 [-1, 64, 32, 32] 64
Upsample-69 [-1, 64, 64, 64] 0
Conv2d-70 [-1, 64, 64, 64] 36,864
_ConvModernNd-71 [-1, 64, 64, 64] 0
_BlockConvStkNd-72 [-1, 64, 64, 64] 0
Conv2d-73 [-1, 3, 64, 63] 4,803
DecoderNet2d-74 [-1, 3, 64, 63] 0
================================================================
Total params: 22,230,531
Trainable params: 22,230,531
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.00
Forward/backward pass size (MB): 19.81
Params size (MB): 84.80
Estimated Total Size (MB): 104.61
----------------------------------------------------------------
Example 2
1 2 3 4 5 6 | |
The number of convolutional layers along the depth is 17.
The input size is (1024, 2, 2).
----------------------------------------------------------------
Layer (type) Output Shape Param #
================================================================
Conv2d-1 [-1, 1024, 2, 2] 9,437,184
InstanceNorm2d-2 [-1, 1024, 2, 2] 2,048
PReLU-3 [-1, 1024, 2, 2] 1,024
Conv2d-4 [-1, 512, 2, 2] 4,718,592
_ConvModernNd-5 [-1, 512, 2, 2] 0
InstanceNorm2d-6 [-1, 512, 2, 2] 1,024
PReLU-7 [-1, 512, 2, 2] 512
Conv2d-8 [-1, 512, 2, 2] 2,359,296
_ConvModernNd-9 [-1, 512, 2, 2] 0
InstanceNorm2d-10 [-1, 512, 2, 2] 1,024
PReLU-11 [-1, 512, 2, 2] 512
Upsample-12 [-1, 512, 4, 4] 0
Conv2d-13 [-1, 512, 4, 4] 2,359,296
_ConvModernNd-14 [-1, 512, 4, 4] 0
_BlockConvStkNd-15 [-1, 512, 4, 4] 0
InstanceNorm2d-16 [-1, 512, 4, 4] 1,024
PReLU-17 [-1, 512, 4, 4] 512
Conv2d-18 [-1, 256, 4, 4] 1,179,648
_ConvModernNd-19 [-1, 256, 4, 4] 0
InstanceNorm2d-20 [-1, 256, 4, 4] 512
PReLU-21 [-1, 256, 4, 4] 256
Conv2d-22 [-1, 256, 4, 4] 589,824
_ConvModernNd-23 [-1, 256, 4, 4] 0
InstanceNorm2d-24 [-1, 256, 4, 4] 512
PReLU-25 [-1, 256, 4, 4] 256
Upsample-26 [-1, 256, 8, 8] 0
Conv2d-27 [-1, 256, 8, 8] 589,824
_ConvModernNd-28 [-1, 256, 8, 8] 0
_BlockConvStkNd-29 [-1, 256, 8, 8] 0
InstanceNorm2d-30 [-1, 256, 8, 8] 512
PReLU-31 [-1, 256, 8, 8] 256
Conv2d-32 [-1, 128, 8, 8] 294,912
_ConvModernNd-33 [-1, 128, 8, 8] 0
InstanceNorm2d-34 [-1, 128, 8, 8] 256
PReLU-35 [-1, 128, 8, 8] 128
Conv2d-36 [-1, 128, 8, 8] 147,456
_ConvModernNd-37 [-1, 128, 8, 8] 0
InstanceNorm2d-38 [-1, 128, 8, 8] 256
PReLU-39 [-1, 128, 8, 8] 128
Upsample-40 [-1, 128, 16, 16] 0
Conv2d-41 [-1, 128, 16, 16] 147,456
_ConvModernNd-42 [-1, 128, 16, 16] 0
_BlockConvStkNd-43 [-1, 128, 16, 16] 0
InstanceNorm2d-44 [-1, 128, 16, 16] 256
PReLU-45 [-1, 128, 16, 16] 128
Conv2d-46 [-1, 64, 16, 16] 73,728
_ConvModernNd-47 [-1, 64, 16, 16] 0
InstanceNorm2d-48 [-1, 64, 16, 16] 128
PReLU-49 [-1, 64, 16, 16] 64
Conv2d-50 [-1, 64, 16, 16] 36,864
_ConvModernNd-51 [-1, 64, 16, 16] 0
InstanceNorm2d-52 [-1, 64, 16, 16] 128
PReLU-53 [-1, 64, 16, 16] 64
Upsample-54 [-1, 64, 32, 32] 0
Conv2d-55 [-1, 64, 32, 32] 36,864
_ConvModernNd-56 [-1, 64, 32, 32] 0
_BlockConvStkNd-57 [-1, 64, 32, 32] 0
InstanceNorm2d-58 [-1, 64, 32, 32] 128
PReLU-59 [-1, 64, 32, 32] 64
Conv2d-60 [-1, 64, 32, 32] 36,864
_ConvModernNd-61 [-1, 64, 32, 32] 0
InstanceNorm2d-62 [-1, 64, 32, 32] 128
PReLU-63 [-1, 64, 32, 32] 64
Conv2d-64 [-1, 64, 32, 32] 36,864
_ConvModernNd-65 [-1, 64, 32, 32] 0
InstanceNorm2d-66 [-1, 64, 32, 32] 128
PReLU-67 [-1, 64, 32, 32] 64
Upsample-68 [-1, 64, 64, 64] 0
Conv2d-69 [-1, 64, 64, 64] 36,864
_ConvModernNd-70 [-1, 64, 64, 64] 0
_BlockConvStkNd-71 [-1, 64, 64, 64] 0
Conv2d-72 [-1, 3, 64, 63] 4,803
DecoderNet2d-73 [-1, 3, 64, 63] 0
================================================================
Total params: 22,098,435
Trainable params: 22,098,435
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.02
Forward/backward pass size (MB): 19.78
Params size (MB): 84.30
Estimated Total Size (MB): 104.09
----------------------------------------------------------------
Last update: March 14, 2021