modules.conv.UNet3d¶
net = mdnc.modules.conv.UNet3d(
channel, layers,
kernel_size=3, in_planes=1, out_planes=1
)
This moule is a built-in model for 3D convolutional U-Net. The network is inspired by:
The network would down-sample and 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
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"]
u1["Block 2n-1<br>Stack of layers[0] layers"]
u2["Block 2n-2<br>Stack of layers[1] layers"]
ui["Block ...<br>Stack of ... layers"]
b1 -->|down<br>sampling| b2 -->|down<br>sampling| bi -->|down<br>sampling| bn
bn -->|up<br>sampling| ui -->|up<br>sampling| u2 -->|up<br>sampling| u1
b1 -->|skip<br>connection| u1
b2 -->|skip<br>connection| u2
bi -->|skip<br>connection| ui
linkStyle 0,1,2 stroke-width:4px, stroke:#800 ;
linkStyle 3,4,5 stroke-width:4px, stroke:#080 ;
linkStyle 6,7,8 stroke-width:4px, stroke:#888 ;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 or up-sampling is configured by stride=2. The channel number would be doubled in the down-sampling route and reduced to ½ in the up-sampling route. The skip connection is perfromed by concatenation.
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. After each up-sampling, the channel number would be reduced to ½. |
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_planes | int | The channel number of the output data. |
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 3D tensor, the size should be (B, C, L1, L2, L3), where B is the batch size, C is the output channel number, and (L1, L2, L3) is the input data size. |
Properties¶
nlayers¶
net.nlayers
The total number of convolutional layers along the depth of the network.
Examples¶
Example
1 2 3 4 5 | |
The number of convolutional layers along the depth is 25.
----------------------------------------------------------------
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, 16, 16, 15] 110,592
_ConvModernNd-9 [-1, 64, 16, 16, 15] 0
_BlockConvStkNd-10 [-1, 64, 16, 16, 15] 0
[-1, 64, 31, 32, 30]
InstanceNorm3d-11 [-1, 64, 16, 16, 15] 128
PReLU-12 [-1, 64, 16, 16, 15] 64
Conv3d-13 [-1, 128, 16, 16, 15] 221,184
_ConvModernNd-14 [-1, 128, 16, 16, 15] 0
InstanceNorm3d-15 [-1, 128, 16, 16, 15] 256
PReLU-16 [-1, 128, 16, 16, 15] 128
Conv3d-17 [-1, 128, 8, 8, 8] 442,368
_ConvModernNd-18 [-1, 128, 8, 8, 8] 0
_BlockConvStkNd-19 [-1, 128, 8, 8, 8] 0
[-1, 128, 16, 16, 15]
InstanceNorm3d-20 [-1, 128, 8, 8, 8] 256
PReLU-21 [-1, 128, 8, 8, 8] 128
Conv3d-22 [-1, 256, 8, 8, 8] 884,736
_ConvModernNd-23 [-1, 256, 8, 8, 8] 0
InstanceNorm3d-24 [-1, 256, 8, 8, 8] 512
PReLU-25 [-1, 256, 8, 8, 8] 256
Conv3d-26 [-1, 256, 8, 8, 8] 1,769,472
_ConvModernNd-27 [-1, 256, 8, 8, 8] 0
InstanceNorm3d-28 [-1, 256, 8, 8, 8] 512
PReLU-29 [-1, 256, 8, 8, 8] 256
Conv3d-30 [-1, 256, 4, 4, 4] 1,769,472
_ConvModernNd-31 [-1, 256, 4, 4, 4] 0
_BlockConvStkNd-32 [-1, 256, 4, 4, 4] 0
[-1, 256, 8, 8, 8]
InstanceNorm3d-33 [-1, 256, 4, 4, 4] 512
PReLU-34 [-1, 256, 4, 4, 4] 256
Conv3d-35 [-1, 512, 4, 4, 4] 3,538,944
_ConvModernNd-36 [-1, 512, 4, 4, 4] 0
InstanceNorm3d-37 [-1, 512, 4, 4, 4] 1,024
PReLU-38 [-1, 512, 4, 4, 4] 512
Conv3d-39 [-1, 512, 4, 4, 4] 7,077,888
_ConvModernNd-40 [-1, 512, 4, 4, 4] 0
InstanceNorm3d-41 [-1, 512, 4, 4, 4] 1,024
PReLU-42 [-1, 512, 4, 4, 4] 512
Conv3d-43 [-1, 512, 2, 2, 2] 7,077,888
_ConvModernNd-44 [-1, 512, 2, 2, 2] 0
_BlockConvStkNd-45 [-1, 512, 2, 2, 2] 0
[-1, 512, 4, 4, 4]
InstanceNorm3d-46 [-1, 512, 2, 2, 2] 1,024
PReLU-47 [-1, 512, 2, 2, 2] 512
Conv3d-48 [-1, 1024, 2, 2, 2] 14,155,776
_ConvModernNd-49 [-1, 1024, 2, 2, 2] 0
InstanceNorm3d-50 [-1, 1024, 2, 2, 2] 2,048
PReLU-51 [-1, 1024, 2, 2, 2] 1,024
Conv3d-52 [-1, 1024, 2, 2, 2] 28,311,552
_ConvModernNd-53 [-1, 1024, 2, 2, 2] 0
InstanceNorm3d-54 [-1, 1024, 2, 2, 2] 2,048
PReLU-55 [-1, 1024, 2, 2, 2] 1,024
Upsample-56 [-1, 1024, 4, 4, 4] 0
Conv3d-57 [-1, 512, 4, 4, 4] 14,155,776
_ConvModernNd-58 [-1, 512, 4, 4, 4] 0
_BlockConvStkNd-59 [-1, 512, 4, 4, 4] 0
InstanceNorm3d-60 [-1, 1024, 4, 4, 4] 2,048
PReLU-61 [-1, 1024, 4, 4, 4] 1,024
Conv3d-62 [-1, 512, 4, 4, 4] 14,155,776
_ConvModernNd-63 [-1, 512, 4, 4, 4] 0
InstanceNorm3d-64 [-1, 512, 4, 4, 4] 1,024
PReLU-65 [-1, 512, 4, 4, 4] 512
Conv3d-66 [-1, 512, 4, 4, 4] 7,077,888
_ConvModernNd-67 [-1, 512, 4, 4, 4] 0
InstanceNorm3d-68 [-1, 512, 4, 4, 4] 1,024
PReLU-69 [-1, 512, 4, 4, 4] 512
Upsample-70 [-1, 512, 8, 8, 8] 0
Conv3d-71 [-1, 256, 8, 8, 8] 3,538,944
_ConvModernNd-72 [-1, 256, 8, 8, 8] 0
_BlockConvStkNd-73 [-1, 256, 8, 8, 8] 0
InstanceNorm3d-74 [-1, 512, 8, 8, 8] 1,024
PReLU-75 [-1, 512, 8, 8, 8] 512
Conv3d-76 [-1, 256, 8, 8, 8] 3,538,944
_ConvModernNd-77 [-1, 256, 8, 8, 8] 0
InstanceNorm3d-78 [-1, 256, 8, 8, 8] 512
PReLU-79 [-1, 256, 8, 8, 8] 256
Conv3d-80 [-1, 256, 8, 8, 8] 1,769,472
_ConvModernNd-81 [-1, 256, 8, 8, 8] 0
InstanceNorm3d-82 [-1, 256, 8, 8, 8] 512
PReLU-83 [-1, 256, 8, 8, 8] 256
Upsample-84 [-1, 256, 16, 16, 16] 0
Conv3d-85 [-1, 128, 16, 16, 16] 884,736
_ConvModernNd-86 [-1, 128, 16, 16, 16] 0
_BlockConvStkNd-87 [-1, 128, 16, 16, 16] 0
InstanceNorm3d-88 [-1, 256, 16, 16, 15] 512
PReLU-89 [-1, 256, 16, 16, 15] 256
Conv3d-90 [-1, 128, 16, 16, 15] 884,736
_ConvModernNd-91 [-1, 128, 16, 16, 15] 0
InstanceNorm3d-92 [-1, 128, 16, 16, 15] 256
PReLU-93 [-1, 128, 16, 16, 15] 128
Upsample-94 [-1, 128, 32, 32, 30] 0
Conv3d-95 [-1, 64, 32, 32, 30] 221,184
_ConvModernNd-96 [-1, 64, 32, 32, 30] 0
_BlockConvStkNd-97 [-1, 64, 32, 32, 30] 0
InstanceNorm3d-98 [-1, 128, 31, 32, 30] 256
PReLU-99 [-1, 128, 31, 32, 30] 128
Conv3d-100 [-1, 64, 31, 32, 30] 221,184
_ConvModernNd-101 [-1, 64, 31, 32, 30] 0
InstanceNorm3d-102 [-1, 64, 31, 32, 30] 128
PReLU-103 [-1, 64, 31, 32, 30] 64
Conv3d-104 [-1, 64, 31, 32, 30] 110,592
_ConvModernNd-105 [-1, 64, 31, 32, 30] 0
_BlockConvStkNd-106 [-1, 64, 31, 32, 30] 0
Conv3d-107 [-1, 1, 31, 32, 30] 8,001
UNet3d-108 [-1, 1, 31, 32, 30] 0
================================================================
Total params: 112,087,041
Trainable params: 112,087,041
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.34
Forward/backward pass size (MB): 465.95
Params size (MB): 427.58
Estimated Total Size (MB): 893.87
----------------------------------------------------------------
Last update: March 14, 2021