create_dl_layer_transposed_convolutionT_create_dl_layer_transposed_convolutionCreateDlLayerTransposedConvolutionCreateDlLayerTransposedConvolutioncreate_dl_layer_transposed_convolution (Operator)

Name

create_dl_layer_transposed_convolutionT_create_dl_layer_transposed_convolutionCreateDlLayerTransposedConvolutionCreateDlLayerTransposedConvolutioncreate_dl_layer_transposed_convolution — Create a transposed convolution layer.

Signature

Description

The operator create_dl_layer_transposed_convolutioncreate_dl_layer_transposed_convolutionCreateDlLayerTransposedConvolutionCreateDlLayerTransposedConvolutionCreateDlLayerTransposedConvolutioncreate_dl_layer_transposed_convolution creates a transposed convolution layer whose handle is returned in DLLayerTransposedConvolutionDLLayerTransposedConvolutionDLLayerTransposedConvolutionDLLayerTransposedConvolutionDLLayerTransposedConvolutiondllayer_transposed_convolution.

The parameter DLLayerInputDLLayerInputDLLayerInputDLLayerInputDLLayerInputdllayer_input determines the feeding input layer and expects the layer handle as value.

The parameter LayerNameLayerNameLayerNameLayerNamelayerNamelayer_name sets an individual layer name. Note that if creating a model using create_dl_modelcreate_dl_modelCreateDlModelCreateDlModelCreateDlModelcreate_dl_model each layer of the created network must have a unique name.

The parameter KernelSizeKernelSizeKernelSizeKernelSizekernelSizekernel_size specifies the filter kernel in the dimensions width and height. So far, only quadratic kernels are supported.

Restriction: This value must be a tuple of length 1.

The parameter StrideStrideStrideStridestridestride determines how the filter is shifted in row and column direction.

Restriction: This value must be a tuple of length 1.

The parameter KernelDepthKernelDepthKernelDepthKernelDepthkernelDepthkernel_depth defines the depth of the output feature maps.

Restriction: This value must be a tuple of length 1.

The parameter GroupsGroupsGroupsGroupsgroupsgroups determines the amount of filter groups. So far, only a single filter group is supported.

Restriction: This value must be a tuple of length 1.

The parameter PaddingPaddingPaddingPaddingpaddingpadding effectively appends pixels with value 0 to each border of the input. This is set so that a convolutional layer and a transposed convolution layer with the same KernelSizeKernelSizeKernelSizeKernelSizekernelSizekernel_size, StrideStrideStrideStridestridestride and PaddingPaddingPaddingPaddingpaddingpadding values are inverses of each other regarding their input and output shapes. Supported PaddingPaddingPaddingPaddingpaddingpadding values are:

• 'half_kernel_size'"half_kernel_size""half_kernel_size""half_kernel_size""half_kernel_size""half_kernel_size": The integer value of in the formula above depends on the specified KernelSizeKernelSizeKernelSizeKernelSizekernelSizekernel_size. More precisely, it is calculated as .

• 'none'"none""none""none""none""none": The value of in the formula above is 0.

• Number of pixels: Specify the integer value of in the formula above for each border. To do so, the following tuple lengths are supported:

  • Single number: value for all four directions left/right/top/bottom.

  • Two numbers: value for left/right and top/bottom: [l/r, t/b].

  • Four numbers: value for left, right, top, bottom side: [l,r,t,b].

    Restriction: 'runtime'"runtime""runtime""runtime""runtime""runtime" 'gpu'"gpu""gpu""gpu""gpu""gpu" does not support asymmetric padding, i.e., the padding values for the left and right side must be equal, as well as the padding values for the top and bottom side.

  Restriction: The integer padding values must be smaller than the value set for KernelSizeKernelSizeKernelSizeKernelSizekernelSizekernel_size.

The following generic parameters GenParamNameGenParamNameGenParamNameGenParamNamegenParamNamegen_param_name and the corresponding values GenParamValueGenParamValueGenParamValueGenParamValuegenParamValuegen_param_value are supported:

'bias_term'"bias_term""bias_term""bias_term""bias_term""bias_term":

Determines, whether the layer has bias terms.

Default: 'true'"true""true""true""true""true"

'is_inference_output'"is_inference_output""is_inference_output""is_inference_output""is_inference_output""is_inference_output":

Determines whether apply_dl_modelapply_dl_modelApplyDlModelApplyDlModelApplyDlModelapply_dl_model will include the output of this layer in the dictionary DLResultBatchDLResultBatchDLResultBatchDLResultBatchDLResultBatchdlresult_batch even without specifying this layer in OutputsOutputsOutputsOutputsoutputsoutputs ('true'"true""true""true""true""true") or not ('false'"false""false""false""false""false").

Default: 'false'"false""false""false""false""false"

'learning_rate_multiplier'"learning_rate_multiplier""learning_rate_multiplier""learning_rate_multiplier""learning_rate_multiplier""learning_rate_multiplier":

Learning rate multiplier for this layer that is used during training. If 'learning_rate_multiplier'"learning_rate_multiplier""learning_rate_multiplier""learning_rate_multiplier""learning_rate_multiplier""learning_rate_multiplier" is set to 0.0, the layer is skipped during training.

Default: 1.0

'output_padding'"output_padding""output_padding""output_padding""output_padding""output_padding":

Can be used to resolve ambiguities in the output shape for KernelSizeKernelSizeKernelSizeKernelSizekernelSizekernel_size and StrideStrideStrideStridestridestride larger than 1. As mentioned above in the description of the parameter PaddingPaddingPaddingPaddingpaddingpadding, with respect to the input and output shapes a transposed convolution layer can be seen as the inverse of a convolution layer if they have the same settings for KernelSizeKernelSizeKernelSizeKernelSizekernelSizekernel_size, StrideStrideStrideStridestridestride, and PaddingPaddingPaddingPaddingpaddingpadding (and dilation). However, a convolution layer can map several shapes to the same output spatial dimensions. E.g., for KernelSizeKernelSizeKernelSizeKernelSizekernelSizekernel_size 3, PaddingPaddingPaddingPaddingpaddingpadding 1, and StrideStrideStrideStridestridestride 2, both inputs with spatial dimensions (H, W) = '(4, 4)'"(4, 4)""(4, 4)""(4, 4)""(4, 4)""(4, 4)" and (H, W) = '(3, 3)'"(3, 3)""(3, 3)""(3, 3)""(3, 3)""(3, 3)" are mapped to an output shape with (H, W) = '(2, 2)'"(2, 2)""(2, 2)""(2, 2)""(2, 2)""(2, 2)". To get back to '(4, 4)'"(4, 4)""(4, 4)""(4, 4)""(4, 4)""(4, 4)" using a transposed convolution (with the same settings for KernelSizeKernelSizeKernelSizeKernelSizekernelSizekernel_size, StrideStrideStrideStridestridestride, and PaddingPaddingPaddingPaddingpaddingpadding) 'output_padding'"output_padding""output_padding""output_padding""output_padding""output_padding" must be set to 1. 'output_padding'"output_padding""output_padding""output_padding""output_padding""output_padding" must not be larger than the bottom and right PaddingPaddingPaddingPaddingpaddingpadding and can only be set larger than 0 if StrideStrideStrideStridestridestride is larger than 1.

Default: 0

'weight_filler'"weight_filler""weight_filler""weight_filler""weight_filler""weight_filler":

Defines the mode how the weights are initialized. See create_dl_layer_convolutioncreate_dl_layer_convolutionCreateDlLayerConvolutionCreateDlLayerConvolutionCreateDlLayerConvolutioncreate_dl_layer_convolution for a detailed explanation of this parameter and its values.

List of values: 'xavier'"xavier""xavier""xavier""xavier""xavier", 'msra'"msra""msra""msra""msra""msra", 'const'"const""const""const""const""const"

Default: 'xavier'"xavier""xavier""xavier""xavier""xavier"

'weight_filler_const_val'"weight_filler_const_val""weight_filler_const_val""weight_filler_const_val""weight_filler_const_val""weight_filler_const_val":

See create_dl_layer_convolutioncreate_dl_layer_convolutionCreateDlLayerConvolutionCreateDlLayerConvolutionCreateDlLayerConvolutioncreate_dl_layer_convolution for a detailed explanation of this parameter and its values.

Default: 0.5

'weight_filler_variance_norm'"weight_filler_variance_norm""weight_filler_variance_norm""weight_filler_variance_norm""weight_filler_variance_norm""weight_filler_variance_norm":

Value range for 'weight_filler'"weight_filler""weight_filler""weight_filler""weight_filler""weight_filler". See create_dl_layer_convolutioncreate_dl_layer_convolutionCreateDlLayerConvolutionCreateDlLayerConvolutionCreateDlLayerConvolutioncreate_dl_layer_convolution for a detailed explanation of this parameter and its values.

List of values: 'norm_average'"norm_average""norm_average""norm_average""norm_average""norm_average", 'norm_in'"norm_in""norm_in""norm_in""norm_in""norm_in", 'norm_out'"norm_out""norm_out""norm_out""norm_out""norm_out", constant value (in combination with 'weight_filler'"weight_filler""weight_filler""weight_filler""weight_filler""weight_filler" = 'msra'"msra""msra""msra""msra""msra")

Default: 'norm_in'"norm_in""norm_in""norm_in""norm_in""norm_in"

Certain parameters of layers created using this operator create_dl_layer_transposed_convolutioncreate_dl_layer_transposed_convolutionCreateDlLayerTransposedConvolutionCreateDlLayerTransposedConvolutionCreateDlLayerTransposedConvolutioncreate_dl_layer_transposed_convolution can be set and retrieved using further operators. The following tables give an overview, which parameters can be set using set_dl_model_layer_paramset_dl_model_layer_paramSetDlModelLayerParamSetDlModelLayerParamSetDlModelLayerParamset_dl_model_layer_param and which ones can be retrieved using get_dl_model_layer_paramget_dl_model_layer_paramGetDlModelLayerParamGetDlModelLayerParamGetDlModelLayerParamget_dl_model_layer_param or get_dl_layer_paramget_dl_layer_paramGetDlLayerParamGetDlLayerParamGetDlLayerParamget_dl_layer_param. Note, the operators set_dl_model_layer_paramset_dl_model_layer_paramSetDlModelLayerParamSetDlModelLayerParamSetDlModelLayerParamset_dl_model_layer_param and get_dl_model_layer_paramget_dl_model_layer_paramGetDlModelLayerParamGetDlModelLayerParamGetDlModelLayerParamget_dl_model_layer_param require a model created by create_dl_modelcreate_dl_modelCreateDlModelCreateDlModelCreateDlModelcreate_dl_model.

Execution Information

• Multithreading type: reentrant (runs in parallel with non-exclusive operators).
• Multithreading scope: global (may be called from any thread).
• Processed without parallelization.

Parameters

Module

Deep Learning Training