modules.resnet.DecoderNet3d¶
net = mdnc.modules.resnet.DecoderNet3d(
channel, layers, out_size, block='bottleneck',
kernel_size=3, in_length=2, out_planes=1
)
This moule is a built-in model for 3D residual 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] blocks"]
u2["Block 2<br>Stack of layers[1] blocks"]
ui["Block ...<br>Stack of ... blocks"]
un["Block n<br>Stack of layers[n-1] blocks"]
optional:::blockoptional
subgraph optional [Optional]
cin["Conv3d<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 residual blocks (see mdnc.modules.resnet.BlockPlain3d and mdnc.modules.resnet.BlockBottleneck3d). Each down-sampling or 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 residual blocks 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, 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. |
block | str | The residual block type, could be:
|
kernel_size | int or(int, int, int) | The kernel size of each residual block. |
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 3D 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 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 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 33.
The input size is (32,).
----------------------------------------------------------------
Layer (type) Output Shape Param #
================================================================
Conv3d-1 [-1, 1024, 1, 1, 1] 33,792
Conv3d-2 [-1, 1024, 1, 1, 1] 28,311,552
InstanceNorm3d-3 [-1, 1024, 1, 1, 1] 2,048
PReLU-4 [-1, 1024, 1, 1, 1] 1,024
Conv3d-5 [-1, 1024, 1, 1, 1] 1,048,576
InstanceNorm3d-6 [-1, 1024, 1, 1, 1] 2,048
PReLU-7 [-1, 1024, 1, 1, 1] 1,024
Conv3d-8 [-1, 1024, 1, 1, 1] 28,311,552
InstanceNorm3d-9 [-1, 1024, 1, 1, 1] 2,048
PReLU-10 [-1, 1024, 1, 1, 1] 1,024
Conv3d-11 [-1, 512, 1, 1, 1] 524,288
Conv3d-12 [-1, 512, 1, 1, 1] 524,288
InstanceNorm3d-13 [-1, 512, 1, 1, 1] 1,024
_BlockBo...neckNd-14 [-1, 512, 1, 1, 1] 0
InstanceNorm3d-15 [-1, 512, 1, 1, 1] 1,024
PReLU-16 [-1, 512, 1, 1, 1] 512
Conv3d-17 [-1, 512, 1, 1, 1] 262,144
InstanceNorm3d-18 [-1, 512, 1, 1, 1] 1,024
PReLU-19 [-1, 512, 1, 1, 1] 512
Upsample-20 [-1, 512, 2, 2, 2] 0
Conv3d-21 [-1, 512, 2, 2, 2] 7,077,888
InstanceNorm3d-22 [-1, 512, 2, 2, 2] 1,024
PReLU-23 [-1, 512, 2, 2, 2] 512
Conv3d-24 [-1, 512, 2, 2, 2] 262,144
Upsample-25 [-1, 512, 2, 2, 2] 0
Conv3d-26 [-1, 512, 2, 2, 2] 262,144
InstanceNorm3d-27 [-1, 512, 2, 2, 2] 1,024
_BlockBo...neckNd-28 [-1, 512, 2, 2, 2] 0
_BlockResStkNd-29 [-1, 512, 2, 2, 2] 0
InstanceNorm3d-30 [-1, 512, 2, 2, 2] 1,024
PReLU-31 [-1, 512, 2, 2, 2] 512
Conv3d-32 [-1, 512, 2, 2, 2] 262,144
InstanceNorm3d-33 [-1, 512, 2, 2, 2] 1,024
PReLU-34 [-1, 512, 2, 2, 2] 512
Conv3d-35 [-1, 512, 2, 2, 2] 7,077,888
InstanceNorm3d-36 [-1, 512, 2, 2, 2] 1,024
PReLU-37 [-1, 512, 2, 2, 2] 512
Conv3d-38 [-1, 256, 2, 2, 2] 131,072
Conv3d-39 [-1, 256, 2, 2, 2] 131,072
InstanceNorm3d-40 [-1, 256, 2, 2, 2] 512
_BlockBo...neckNd-41 [-1, 256, 2, 2, 2] 0
InstanceNorm3d-42 [-1, 256, 2, 2, 2] 512
PReLU-43 [-1, 256, 2, 2, 2] 256
Conv3d-44 [-1, 256, 2, 2, 2] 65,536
InstanceNorm3d-45 [-1, 256, 2, 2, 2] 512
PReLU-46 [-1, 256, 2, 2, 2] 256
Upsample-47 [-1, 256, 4, 4, 4] 0
Conv3d-48 [-1, 256, 4, 4, 4] 1,769,472
InstanceNorm3d-49 [-1, 256, 4, 4, 4] 512
PReLU-50 [-1, 256, 4, 4, 4] 256
Conv3d-51 [-1, 256, 4, 4, 4] 65,536
Upsample-52 [-1, 256, 4, 4, 4] 0
Conv3d-53 [-1, 256, 4, 4, 4] 65,536
InstanceNorm3d-54 [-1, 256, 4, 4, 4] 512
_BlockBo...neckNd-55 [-1, 256, 4, 4, 4] 0
_BlockResStkNd-56 [-1, 256, 4, 4, 4] 0
InstanceNorm3d-57 [-1, 256, 4, 4, 4] 512
PReLU-58 [-1, 256, 4, 4, 4] 256
Conv3d-59 [-1, 256, 4, 4, 4] 65,536
InstanceNorm3d-60 [-1, 256, 4, 4, 4] 512
PReLU-61 [-1, 256, 4, 4, 4] 256
Conv3d-62 [-1, 256, 4, 4, 4] 1,769,472
InstanceNorm3d-63 [-1, 256, 4, 4, 4] 512
PReLU-64 [-1, 256, 4, 4, 4] 256
Conv3d-65 [-1, 128, 4, 4, 4] 32,768
Conv3d-66 [-1, 128, 4, 4, 4] 32,768
InstanceNorm3d-67 [-1, 128, 4, 4, 4] 256
_BlockBo...neckNd-68 [-1, 128, 4, 4, 4] 0
InstanceNorm3d-69 [-1, 128, 4, 4, 4] 256
PReLU-70 [-1, 128, 4, 4, 4] 128
Conv3d-71 [-1, 128, 4, 4, 4] 16,384
InstanceNorm3d-72 [-1, 128, 4, 4, 4] 256
PReLU-73 [-1, 128, 4, 4, 4] 128
Upsample-74 [-1, 128, 8, 8, 8] 0
Conv3d-75 [-1, 128, 8, 8, 8] 442,368
InstanceNorm3d-76 [-1, 128, 8, 8, 8] 256
PReLU-77 [-1, 128, 8, 8, 8] 128
Conv3d-78 [-1, 128, 8, 8, 8] 16,384
Upsample-79 [-1, 128, 8, 8, 8] 0
Conv3d-80 [-1, 128, 8, 8, 8] 16,384
InstanceNorm3d-81 [-1, 128, 8, 8, 8] 256
_BlockBo...neckNd-82 [-1, 128, 8, 8, 8] 0
_BlockResStkNd-83 [-1, 128, 8, 8, 8] 0
InstanceNorm3d-84 [-1, 128, 8, 8, 8] 256
PReLU-85 [-1, 128, 8, 8, 8] 128
Conv3d-86 [-1, 128, 8, 8, 8] 16,384
InstanceNorm3d-87 [-1, 128, 8, 8, 8] 256
PReLU-88 [-1, 128, 8, 8, 8] 128
Conv3d-89 [-1, 128, 8, 8, 8] 442,368
InstanceNorm3d-90 [-1, 128, 8, 8, 8] 256
PReLU-91 [-1, 128, 8, 8, 8] 128
Conv3d-92 [-1, 64, 8, 8, 8] 8,192
Conv3d-93 [-1, 64, 8, 8, 8] 8,192
InstanceNorm3d-94 [-1, 64, 8, 8, 8] 128
_BlockBo...neckNd-95 [-1, 64, 8, 8, 8] 0
InstanceNorm3d-96 [-1, 64, 8, 8, 8] 128
PReLU-97 [-1, 64, 8, 8, 8] 64
Conv3d-98 [-1, 64, 8, 8, 8] 4,096
InstanceNorm3d-99 [-1, 64, 8, 8, 8] 128
PReLU-100 [-1, 64, 8, 8, 8] 64
Upsample-101 [-1, 64, 16, 16, 16] 0
Conv3d-102 [-1, 64, 16, 16, 16] 110,592
InstanceNorm3d-103 [-1, 64, 16, 16, 16] 128
PReLU-104 [-1, 64, 16, 16, 16] 64
Conv3d-105 [-1, 64, 16, 16, 16] 4,096
Upsample-106 [-1, 64, 16, 16, 16] 0
Conv3d-107 [-1, 64, 16, 16, 16] 4,096
InstanceNorm3d-108 [-1, 64, 16, 16, 16] 128
_BlockBo...eckNd-109 [-1, 64, 16, 16, 16] 0
_BlockResStkNd-110 [-1, 64, 16, 16, 16] 0
InstanceNorm3d-111 [-1, 64, 16, 16, 15] 128
PReLU-112 [-1, 64, 16, 16, 15] 64
Conv3d-113 [-1, 64, 16, 16, 15] 4,096
InstanceNorm3d-114 [-1, 64, 16, 16, 15] 128
PReLU-115 [-1, 64, 16, 16, 15] 64
Conv3d-116 [-1, 64, 16, 16, 15] 110,592
InstanceNorm3d-117 [-1, 64, 16, 16, 15] 128
PReLU-118 [-1, 64, 16, 16, 15] 64
Conv3d-119 [-1, 64, 16, 16, 15] 4,096
_BlockBo...eckNd-120 [-1, 64, 16, 16, 15] 0
InstanceNorm3d-121 [-1, 64, 16, 16, 15] 128
PReLU-122 [-1, 64, 16, 16, 15] 64
Conv3d-123 [-1, 64, 16, 16, 15] 4,096
InstanceNorm3d-124 [-1, 64, 16, 16, 15] 128
PReLU-125 [-1, 64, 16, 16, 15] 64
Upsample-126 [-1, 64, 32, 32, 30] 0
Conv3d-127 [-1, 64, 32, 32, 30] 110,592
InstanceNorm3d-128 [-1, 64, 32, 32, 30] 128
PReLU-129 [-1, 64, 32, 32, 30] 64
Conv3d-130 [-1, 64, 32, 32, 30] 4,096
Upsample-131 [-1, 64, 32, 32, 30] 0
Conv3d-132 [-1, 64, 32, 32, 30] 4,096
InstanceNorm3d-133 [-1, 64, 32, 32, 30] 128
_BlockBo...eckNd-134 [-1, 64, 32, 32, 30] 0
_BlockResStkNd-135 [-1, 64, 32, 32, 30] 0
Conv3d-136 [-1, 3, 31, 32, 30] 24,003
DecoderNet3d-137 [-1, 3, 31, 32, 30] 0
================================================================
Total params: 79,473,411
Trainable params: 79,473,411
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.00
Forward/backward pass size (MB): 214.37
Params size (MB): 303.17
Estimated Total Size (MB): 517.53
----------------------------------------------------------------
Example 2
1 2 3 4 5 6 | |
The number of convolutional layers along the depth is 32.
The input size is (1024, 1, 1, 1).
----------------------------------------------------------------
Layer (type) Output Shape Param #
================================================================
Conv3d-1 [-1, 1024, 1, 1, 1] 28,311,552
InstanceNorm3d-2 [-1, 1024, 1, 1, 1] 2,048
PReLU-3 [-1, 1024, 1, 1, 1] 1,024
Conv3d-4 [-1, 1024, 1, 1, 1] 1,048,576
InstanceNorm3d-5 [-1, 1024, 1, 1, 1] 2,048
PReLU-6 [-1, 1024, 1, 1, 1] 1,024
Conv3d-7 [-1, 1024, 1, 1, 1] 28,311,552
InstanceNorm3d-8 [-1, 1024, 1, 1, 1] 2,048
PReLU-9 [-1, 1024, 1, 1, 1] 1,024
Conv3d-10 [-1, 512, 1, 1, 1] 524,288
Conv3d-11 [-1, 512, 1, 1, 1] 524,288
InstanceNorm3d-12 [-1, 512, 1, 1, 1] 1,024
_BlockBo...neckNd-13 [-1, 512, 1, 1, 1] 0
InstanceNorm3d-14 [-1, 512, 1, 1, 1] 1,024
PReLU-15 [-1, 512, 1, 1, 1] 512
Conv3d-16 [-1, 512, 1, 1, 1] 262,144
InstanceNorm3d-17 [-1, 512, 1, 1, 1] 1,024
PReLU-18 [-1, 512, 1, 1, 1] 512
Upsample-19 [-1, 512, 2, 2, 2] 0
Conv3d-20 [-1, 512, 2, 2, 2] 7,077,888
InstanceNorm3d-21 [-1, 512, 2, 2, 2] 1,024
PReLU-22 [-1, 512, 2, 2, 2] 512
Conv3d-23 [-1, 512, 2, 2, 2] 262,144
Upsample-24 [-1, 512, 2, 2, 2] 0
Conv3d-25 [-1, 512, 2, 2, 2] 262,144
InstanceNorm3d-26 [-1, 512, 2, 2, 2] 1,024
_BlockBo...neckNd-27 [-1, 512, 2, 2, 2] 0
_BlockResStkNd-28 [-1, 512, 2, 2, 2] 0
InstanceNorm3d-29 [-1, 512, 2, 2, 2] 1,024
PReLU-30 [-1, 512, 2, 2, 2] 512
Conv3d-31 [-1, 512, 2, 2, 2] 262,144
InstanceNorm3d-32 [-1, 512, 2, 2, 2] 1,024
PReLU-33 [-1, 512, 2, 2, 2] 512
Conv3d-34 [-1, 512, 2, 2, 2] 7,077,888
InstanceNorm3d-35 [-1, 512, 2, 2, 2] 1,024
PReLU-36 [-1, 512, 2, 2, 2] 512
Conv3d-37 [-1, 256, 2, 2, 2] 131,072
Conv3d-38 [-1, 256, 2, 2, 2] 131,072
InstanceNorm3d-39 [-1, 256, 2, 2, 2] 512
_BlockBo...neckNd-40 [-1, 256, 2, 2, 2] 0
InstanceNorm3d-41 [-1, 256, 2, 2, 2] 512
PReLU-42 [-1, 256, 2, 2, 2] 256
Conv3d-43 [-1, 256, 2, 2, 2] 65,536
InstanceNorm3d-44 [-1, 256, 2, 2, 2] 512
PReLU-45 [-1, 256, 2, 2, 2] 256
Upsample-46 [-1, 256, 4, 4, 4] 0
Conv3d-47 [-1, 256, 4, 4, 4] 1,769,472
InstanceNorm3d-48 [-1, 256, 4, 4, 4] 512
PReLU-49 [-1, 256, 4, 4, 4] 256
Conv3d-50 [-1, 256, 4, 4, 4] 65,536
Upsample-51 [-1, 256, 4, 4, 4] 0
Conv3d-52 [-1, 256, 4, 4, 4] 65,536
InstanceNorm3d-53 [-1, 256, 4, 4, 4] 512
_BlockBo...neckNd-54 [-1, 256, 4, 4, 4] 0
_BlockResStkNd-55 [-1, 256, 4, 4, 4] 0
InstanceNorm3d-56 [-1, 256, 4, 4, 4] 512
PReLU-57 [-1, 256, 4, 4, 4] 256
Conv3d-58 [-1, 256, 4, 4, 4] 65,536
InstanceNorm3d-59 [-1, 256, 4, 4, 4] 512
PReLU-60 [-1, 256, 4, 4, 4] 256
Conv3d-61 [-1, 256, 4, 4, 4] 1,769,472
InstanceNorm3d-62 [-1, 256, 4, 4, 4] 512
PReLU-63 [-1, 256, 4, 4, 4] 256
Conv3d-64 [-1, 128, 4, 4, 4] 32,768
Conv3d-65 [-1, 128, 4, 4, 4] 32,768
InstanceNorm3d-66 [-1, 128, 4, 4, 4] 256
_BlockBo...neckNd-67 [-1, 128, 4, 4, 4] 0
InstanceNorm3d-68 [-1, 128, 4, 4, 4] 256
PReLU-69 [-1, 128, 4, 4, 4] 128
Conv3d-70 [-1, 128, 4, 4, 4] 16,384
InstanceNorm3d-71 [-1, 128, 4, 4, 4] 256
PReLU-72 [-1, 128, 4, 4, 4] 128
Upsample-73 [-1, 128, 8, 8, 8] 0
Conv3d-74 [-1, 128, 8, 8, 8] 442,368
InstanceNorm3d-75 [-1, 128, 8, 8, 8] 256
PReLU-76 [-1, 128, 8, 8, 8] 128
Conv3d-77 [-1, 128, 8, 8, 8] 16,384
Upsample-78 [-1, 128, 8, 8, 8] 0
Conv3d-79 [-1, 128, 8, 8, 8] 16,384
InstanceNorm3d-80 [-1, 128, 8, 8, 8] 256
_BlockBo...neckNd-81 [-1, 128, 8, 8, 8] 0
_BlockResStkNd-82 [-1, 128, 8, 8, 8] 0
InstanceNorm3d-83 [-1, 128, 8, 8, 8] 256
PReLU-84 [-1, 128, 8, 8, 8] 128
Conv3d-85 [-1, 128, 8, 8, 8] 16,384
InstanceNorm3d-86 [-1, 128, 8, 8, 8] 256
PReLU-87 [-1, 128, 8, 8, 8] 128
Conv3d-88 [-1, 128, 8, 8, 8] 442,368
InstanceNorm3d-89 [-1, 128, 8, 8, 8] 256
PReLU-90 [-1, 128, 8, 8, 8] 128
Conv3d-91 [-1, 64, 8, 8, 8] 8,192
Conv3d-92 [-1, 64, 8, 8, 8] 8,192
InstanceNorm3d-93 [-1, 64, 8, 8, 8] 128
_BlockBo...neckNd-94 [-1, 64, 8, 8, 8] 0
InstanceNorm3d-95 [-1, 64, 8, 8, 8] 128
PReLU-96 [-1, 64, 8, 8, 8] 64
Conv3d-97 [-1, 64, 8, 8, 8] 4,096
InstanceNorm3d-98 [-1, 64, 8, 8, 8] 128
PReLU-99 [-1, 64, 8, 8, 8] 64
Upsample-100 [-1, 64, 16, 16, 16] 0
Conv3d-101 [-1, 64, 16, 16, 16] 110,592
InstanceNorm3d-102 [-1, 64, 16, 16, 16] 128
PReLU-103 [-1, 64, 16, 16, 16] 64
Conv3d-104 [-1, 64, 16, 16, 16] 4,096
Upsample-105 [-1, 64, 16, 16, 16] 0
Conv3d-106 [-1, 64, 16, 16, 16] 4,096
InstanceNorm3d-107 [-1, 64, 16, 16, 16] 128
_BlockBo...eckNd-108 [-1, 64, 16, 16, 16] 0
_BlockResStkNd-109 [-1, 64, 16, 16, 16] 0
InstanceNorm3d-110 [-1, 64, 16, 16, 15] 128
PReLU-111 [-1, 64, 16, 16, 15] 64
Conv3d-112 [-1, 64, 16, 16, 15] 4,096
InstanceNorm3d-113 [-1, 64, 16, 16, 15] 128
PReLU-114 [-1, 64, 16, 16, 15] 64
Conv3d-115 [-1, 64, 16, 16, 15] 110,592
InstanceNorm3d-116 [-1, 64, 16, 16, 15] 128
PReLU-117 [-1, 64, 16, 16, 15] 64
Conv3d-118 [-1, 64, 16, 16, 15] 4,096
_BlockBo...eckNd-119 [-1, 64, 16, 16, 15] 0
InstanceNorm3d-120 [-1, 64, 16, 16, 15] 128
PReLU-121 [-1, 64, 16, 16, 15] 64
Conv3d-122 [-1, 64, 16, 16, 15] 4,096
InstanceNorm3d-123 [-1, 64, 16, 16, 15] 128
PReLU-124 [-1, 64, 16, 16, 15] 64
Upsample-125 [-1, 64, 32, 32, 30] 0
Conv3d-126 [-1, 64, 32, 32, 30] 110,592
InstanceNorm3d-127 [-1, 64, 32, 32, 30] 128
PReLU-128 [-1, 64, 32, 32, 30] 64
Conv3d-129 [-1, 64, 32, 32, 30] 4,096
Upsample-130 [-1, 64, 32, 32, 30] 0
Conv3d-131 [-1, 64, 32, 32, 30] 4,096
InstanceNorm3d-132 [-1, 64, 32, 32, 30] 128
_BlockBo...eckNd-133 [-1, 64, 32, 32, 30] 0
_BlockResStkNd-134 [-1, 64, 32, 32, 30] 0
Conv3d-135 [-1, 3, 31, 32, 30] 24,003
DecoderNet3d-136 [-1, 3, 31, 32, 30] 0
================================================================
Total params: 79,439,619
Trainable params: 79,439,619
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.00
Forward/backward pass size (MB): 214.36
Params size (MB): 303.04
Estimated Total Size (MB): 517.40
----------------------------------------------------------------
Last update: March 14, 2021