modules.conv.EncoderNet3d¶
net = mdnc.modules.conv.EncoderNet3d(
channel, layers,
kernel_size=3, in_planes=1, out_length=2
)
This moule is a built-in model for 3D convolutional encoder network. This network could be used as a part of the auto-encoder, or just a network for down-sampling data.
The network would down-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
b1["Block 1<br>Stack of layers[0] layers"]
b2["Block 2<br>Stack of layers[1] layers"]
bi["Block ...<br>Stack of ... layers"]
bn["Block n<br>Stack of layers[n-1] layers"]
optional:::blockoptional
subgraph optional [Optional]
fc["FC layer"]
end
b1 -->|down<br>sampling| b2 -->|down<br>sampling| bi -->|down<br>sampling| bn
bn -.->|flatten| fc
linkStyle 0,1,2 stroke-width:4px, stroke:#800 ;
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.ConvModern3d). Each down-sampling is configured by stride=2. The channel number would be doubled in the down-sampling route. An optional flattener and fully-connected layer could be appended to the last layer when the argument out_length != None.
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. |
kernel_size | int or(int, int, int) | The kernel size of each convolutional layer. |
in_planes | int | The channel number of the input data. |
out_length | int | The length of the output vector, if not set, the output would not be flattened. |
Operators¶
__call__¶
y = net(x)
The forward operator implemented by the forward() method. The input is a 3D tensor, and the output is the final output of this network.
Requries
| Argument | Type | Description |
|---|---|---|
x | torch.Tensor | A 3D tensor, the size should be (B, C, L1, L2, L3), where B is the batch size, C is the input channel number, and (L1, L2, L3) is the input data size. |
Returns
| Argument | Description |
|---|---|
y | A tensor,
|
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.
Examples¶
Example 1
1 2 3 4 5 | |
The number of convolutional layers along the depth is 17.
----------------------------------------------------------------
Layer (type) Output Shape Param #
================================================================
Conv3d-1 [-1, 64, 31, 32, 30] 24,000
InstanceNorm3d-2 [-1, 64, 31, 32, 30] 128
PReLU-3 [-1, 64, 31, 32, 30] 64
Conv3d-4 [-1, 64, 31, 32, 30] 110,592
_ConvModernNd-5 [-1, 64, 31, 32, 30] 0
InstanceNorm3d-6 [-1, 64, 31, 32, 30] 128
PReLU-7 [-1, 64, 31, 32, 30] 64
Conv3d-8 [-1, 64, 31, 32, 30] 110,592
_ConvModernNd-9 [-1, 64, 31, 32, 30] 0
InstanceNorm3d-10 [-1, 64, 31, 32, 30] 128
PReLU-11 [-1, 64, 31, 32, 30] 64
Conv3d-12 [-1, 64, 16, 16, 15] 110,592
_ConvModernNd-13 [-1, 64, 16, 16, 15] 0
_BlockConvStkNd-14 [-1, 64, 16, 16, 15] 0
InstanceNorm3d-15 [-1, 64, 16, 16, 15] 128
PReLU-16 [-1, 64, 16, 16, 15] 64
Conv3d-17 [-1, 128, 16, 16, 15] 221,184
_ConvModernNd-18 [-1, 128, 16, 16, 15] 0
InstanceNorm3d-19 [-1, 128, 16, 16, 15] 256
PReLU-20 [-1, 128, 16, 16, 15] 128
Conv3d-21 [-1, 128, 16, 16, 15] 442,368
_ConvModernNd-22 [-1, 128, 16, 16, 15] 0
InstanceNorm3d-23 [-1, 128, 16, 16, 15] 256
PReLU-24 [-1, 128, 16, 16, 15] 128
Conv3d-25 [-1, 128, 8, 8, 8] 442,368
_ConvModernNd-26 [-1, 128, 8, 8, 8] 0
_BlockConvStkNd-27 [-1, 128, 8, 8, 8] 0
InstanceNorm3d-28 [-1, 128, 8, 8, 8] 256
PReLU-29 [-1, 128, 8, 8, 8] 128
Conv3d-30 [-1, 256, 8, 8, 8] 884,736
_ConvModernNd-31 [-1, 256, 8, 8, 8] 0
InstanceNorm3d-32 [-1, 256, 8, 8, 8] 512
PReLU-33 [-1, 256, 8, 8, 8] 256
Conv3d-34 [-1, 256, 8, 8, 8] 1,769,472
_ConvModernNd-35 [-1, 256, 8, 8, 8] 0
InstanceNorm3d-36 [-1, 256, 8, 8, 8] 512
PReLU-37 [-1, 256, 8, 8, 8] 256
Conv3d-38 [-1, 256, 4, 4, 4] 1,769,472
_ConvModernNd-39 [-1, 256, 4, 4, 4] 0
_BlockConvStkNd-40 [-1, 256, 4, 4, 4] 0
InstanceNorm3d-41 [-1, 256, 4, 4, 4] 512
PReLU-42 [-1, 256, 4, 4, 4] 256
Conv3d-43 [-1, 512, 4, 4, 4] 3,538,944
_ConvModernNd-44 [-1, 512, 4, 4, 4] 0
InstanceNorm3d-45 [-1, 512, 4, 4, 4] 1,024
PReLU-46 [-1, 512, 4, 4, 4] 512
Conv3d-47 [-1, 512, 4, 4, 4] 7,077,888
_ConvModernNd-48 [-1, 512, 4, 4, 4] 0
InstanceNorm3d-49 [-1, 512, 4, 4, 4] 1,024
PReLU-50 [-1, 512, 4, 4, 4] 512
Conv3d-51 [-1, 512, 2, 2, 2] 7,077,888
_ConvModernNd-52 [-1, 512, 2, 2, 2] 0
_BlockConvStkNd-53 [-1, 512, 2, 2, 2] 0
InstanceNorm3d-54 [-1, 512, 2, 2, 2] 1,024
PReLU-55 [-1, 512, 2, 2, 2] 512
Conv3d-56 [-1, 1024, 2, 2, 2] 14,155,776
_ConvModernNd-57 [-1, 1024, 2, 2, 2] 0
InstanceNorm3d-58 [-1, 1024, 2, 2, 2] 2,048
PReLU-59 [-1, 1024, 2, 2, 2] 1,024
Conv3d-60 [-1, 1024, 2, 2, 2] 28,311,552
_ConvModernNd-61 [-1, 1024, 2, 2, 2] 0
InstanceNorm3d-62 [-1, 1024, 2, 2, 2] 2,048
PReLU-63 [-1, 1024, 2, 2, 2] 1,024
Conv3d-64 [-1, 1024, 1, 1, 1] 28,311,552
_ConvModernNd-65 [-1, 1024, 1, 1, 1] 0
_BlockConvStkNd-66 [-1, 1024, 1, 1, 1] 0
Conv3d-67 [-1, 1024, 1, 1, 1] 28,312,576
AdaptiveAvgPool3d-68 [-1, 1024, 1, 1, 1] 0
Linear-69 [-1, 32] 32,800
EncoderNet3d-70 [-1, 32] 0
================================================================
Total params: 122,719,328
Trainable params: 122,719,328
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.34
Forward/backward pass size (MB): 213.04
Params size (MB): 468.14
Estimated Total Size (MB): 681.52
----------------------------------------------------------------
Example 2
1 2 3 4 5 | |
The number of convolutional layers along the depth is 17.
----------------------------------------------------------------
Layer (type) Output Shape Param #
================================================================
Conv3d-1 [-1, 64, 31, 32, 30] 24,000
InstanceNorm3d-2 [-1, 64, 31, 32, 30] 128
PReLU-3 [-1, 64, 31, 32, 30] 64
Conv3d-4 [-1, 64, 31, 32, 30] 110,592
_ConvModernNd-5 [-1, 64, 31, 32, 30] 0
InstanceNorm3d-6 [-1, 64, 31, 32, 30] 128
PReLU-7 [-1, 64, 31, 32, 30] 64
Conv3d-8 [-1, 64, 31, 32, 30] 110,592
_ConvModernNd-9 [-1, 64, 31, 32, 30] 0
InstanceNorm3d-10 [-1, 64, 31, 32, 30] 128
PReLU-11 [-1, 64, 31, 32, 30] 64
Conv3d-12 [-1, 64, 16, 16, 15] 110,592
_ConvModernNd-13 [-1, 64, 16, 16, 15] 0
_BlockConvStkNd-14 [-1, 64, 16, 16, 15] 0
InstanceNorm3d-15 [-1, 64, 16, 16, 15] 128
PReLU-16 [-1, 64, 16, 16, 15] 64
Conv3d-17 [-1, 128, 16, 16, 15] 221,184
_ConvModernNd-18 [-1, 128, 16, 16, 15] 0
InstanceNorm3d-19 [-1, 128, 16, 16, 15] 256
PReLU-20 [-1, 128, 16, 16, 15] 128
Conv3d-21 [-1, 128, 16, 16, 15] 442,368
_ConvModernNd-22 [-1, 128, 16, 16, 15] 0
InstanceNorm3d-23 [-1, 128, 16, 16, 15] 256
PReLU-24 [-1, 128, 16, 16, 15] 128
Conv3d-25 [-1, 128, 8, 8, 8] 442,368
_ConvModernNd-26 [-1, 128, 8, 8, 8] 0
_BlockConvStkNd-27 [-1, 128, 8, 8, 8] 0
InstanceNorm3d-28 [-1, 128, 8, 8, 8] 256
PReLU-29 [-1, 128, 8, 8, 8] 128
Conv3d-30 [-1, 256, 8, 8, 8] 884,736
_ConvModernNd-31 [-1, 256, 8, 8, 8] 0
InstanceNorm3d-32 [-1, 256, 8, 8, 8] 512
PReLU-33 [-1, 256, 8, 8, 8] 256
Conv3d-34 [-1, 256, 8, 8, 8] 1,769,472
_ConvModernNd-35 [-1, 256, 8, 8, 8] 0
InstanceNorm3d-36 [-1, 256, 8, 8, 8] 512
PReLU-37 [-1, 256, 8, 8, 8] 256
Conv3d-38 [-1, 256, 4, 4, 4] 1,769,472
_ConvModernNd-39 [-1, 256, 4, 4, 4] 0
_BlockConvStkNd-40 [-1, 256, 4, 4, 4] 0
InstanceNorm3d-41 [-1, 256, 4, 4, 4] 512
PReLU-42 [-1, 256, 4, 4, 4] 256
Conv3d-43 [-1, 512, 4, 4, 4] 3,538,944
_ConvModernNd-44 [-1, 512, 4, 4, 4] 0
InstanceNorm3d-45 [-1, 512, 4, 4, 4] 1,024
PReLU-46 [-1, 512, 4, 4, 4] 512
Conv3d-47 [-1, 512, 4, 4, 4] 7,077,888
_ConvModernNd-48 [-1, 512, 4, 4, 4] 0
InstanceNorm3d-49 [-1, 512, 4, 4, 4] 1,024
PReLU-50 [-1, 512, 4, 4, 4] 512
Conv3d-51 [-1, 512, 2, 2, 2] 7,077,888
_ConvModernNd-52 [-1, 512, 2, 2, 2] 0
_BlockConvStkNd-53 [-1, 512, 2, 2, 2] 0
InstanceNorm3d-54 [-1, 512, 2, 2, 2] 1,024
PReLU-55 [-1, 512, 2, 2, 2] 512
Conv3d-56 [-1, 1024, 2, 2, 2] 14,155,776
_ConvModernNd-57 [-1, 1024, 2, 2, 2] 0
InstanceNorm3d-58 [-1, 1024, 2, 2, 2] 2,048
PReLU-59 [-1, 1024, 2, 2, 2] 1,024
Conv3d-60 [-1, 1024, 2, 2, 2] 28,311,552
_ConvModernNd-61 [-1, 1024, 2, 2, 2] 0
InstanceNorm3d-62 [-1, 1024, 2, 2, 2] 2,048
PReLU-63 [-1, 1024, 2, 2, 2] 1,024
Conv3d-64 [-1, 1024, 1, 1, 1] 28,311,552
_ConvModernNd-65 [-1, 1024, 1, 1, 1] 0
_BlockConvStkNd-66 [-1, 1024, 1, 1, 1] 0
Conv3d-67 [-1, 1024, 1, 1, 1] 28,312,576
EncoderNet3d-68 [-1, 1024, 1, 1, 1] 0
================================================================
Total params: 122,686,528
Trainable params: 122,686,528
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.34
Forward/backward pass size (MB): 213.04
Params size (MB): 468.01
Estimated Total Size (MB): 681.39
----------------------------------------------------------------
Last update: March 14, 2021