create_dl_layer_slice🔗
Short description🔗
create_dl_layer_slice — Create a slice layer.
Signature🔗
create_dl_layer_slice( dl_layer DLLayerInput, string LayerName, number Axes, number StartIndices, number EndIndices, number Steps, attribute.name GenParamName, attribute.value GenParamValue, out dl_layer DLLayerSlice )
Description🔗
The operator create_dl_layer_slice creates a slice layer that extracts
a single slice of a tensor along specified axes, based on the provided start
and end indices as well as the step size for the slicing operations.
The handle of the slice layer is returned in DLLayerSlice.
The parameter DLLayerInput determines the feeding input layer and
expects the layer handle as value.
The parameter LayerName sets an individual layer name.
Note that if creating a model using create_dl_model each layer of
the created network must have a unique name.
The following parameters are used for the slice operation:
-
'starts' Determines the starting indices of the slice operation along the axes specified in 'axes'. This parameter is set using
StartIndices. -
'ends' Determines the ending indices of the slice operation along the axes specified in 'axes'. This parameter is set using
EndIndices. -
'steps' Determines the step size for the slicing operation along the axes specified in 'axes'. The parameter 'steps' must either be an empty tuple or have the same length as 'axes'. If it is empty, the 'steps' are set to 1 for all 'axes'. This parameter is set using
Steps. -
'axes' Specifies the axes along which the slicing operation is performed. The axes are defined in the format WHC, representing Width, Height, and Channel and can also be specified as strings. Slicing in the Batch dimension is not supported. This parameter is set using
Axes.
The slice layer allows slicing a tensor along specified axes, providing flexible adjustment options. To ensure correct operations 'starts' and 'ends' must have the same length as the parameter 'axes'. Negative values in 'starts' and 'ends' are adjusted by adding the size of the corresponding array dimension, effectively allowing negative indexing for one iteration. After that, negative values are clamped to prevent further out-of-bounds slicing.
Negative values in 'axes' are adjusted by adding the total number of dimensions of the tensor to the respective axis. Negative value means counting dimensions from the back. The accepted range is [\(-r\), \(r-1\)] where \(r\) = rank. For slicing to the end of a dimension with unknown size, it is recommended to pass in H_INT_MAX when slicing forward and H_INT_MIN when slicing backward.
Handling of axes and indices
-
Adjusting axes Negative values in the 'axes' parameter are made non-negative by adding the rank of the input tensor \(r\) to each negative axis:
\[\begin{eqnarray*} \textrm{axes}[i] \gets \textrm{axes}[i] + r \quad \textrm{if } \textrm{axes}[i] < 0 \end{eqnarray*}\] -
Adjusting ‘starts’ and ‘ends’
-
Negative Index Adjustment Negative values in 'starts' and 'ends' are adjusted by adding the respective tensor dimension size:
'starts'[\(i\)] is adjusted by adding the tensor dimension size.
\[\begin{eqnarray*} \textrm{starts}[i] \gets \textrm{starts}[i] + \textrm{dims}[\textrm{axes}[i]] \quad \textrm{if } \textrm{starts}[i] < 0 \end{eqnarray*}\]'ends'[\(i\)] is adjusted by adding the respective tensor dimension size.
\[\begin{eqnarray*} \textrm{ends}[i] \gets \textrm{ends}[i] + \textrm{dims}[\textrm{axes}[i]] \quad \textrm{if } \textrm{ends}[i] < 0 \end{eqnarray*}\] -
Clamping ‘starts’ Values Adjusted 'starts' values are clamped to the range:
\[\begin{eqnarray*} \textrm{starts}[i] \in [0, \textrm{dims}[\textrm{axes}[i]]] \end{eqnarray*}\] -
Clamping ‘ends’ Values Adjusted 'ends' values are clamped based on the step direction:
-
Positive Stepping: 'ends' is clamped to the range [0, dims].
\[\begin{eqnarray*} \textrm{if } \textrm{steps}[i] > 0: \, \textrm{ends}[i] \in [0, \textrm{dims}[\textrm{axes}[i]]] \end{eqnarray*}\] -
Negative Stepping: 'ends' is clamped to the range [-1, dims - 1].
\[\begin{eqnarray*} \textrm{if } \textrm{steps}[i] < 0: \, \textrm{ends}[i] \in [-1, \textrm{dims}[\textrm{axes}[i]] - 1] \end{eqnarray*}\]
-
-
The following generic parameters GenParamName and the corresponding
values GenParamValue are supported:
-
'axes_format': Compatibility parameter that determines whether the specified
Axesare given in WHCN format (default) or in NCHW format. The parameter changes how the specified axes are interpreted. For example, with axis 1 and 'axes_format' 'nchw', slicing is performed along depth. If 'axes_format' is set to 'whcn' or left unspecified, slicing is performed along height. Instead of indices, strings can also be used to specify the axes.List of values: 'nchw', 'whcn'
Default: 'whcn'
-
'is_inference_output': Determines whether
apply_dl_modelwill include the output of this layer in the dictionaryDLResultBatcheven without specifying this layer inOutputs('true') or not ('false').List of values: 'true', 'false'
Default: 'false'
Certain parameters of layers created using this operator
create_dl_layer_slice can be set and retrieved using
further operators.
The following tables give an overview, which parameters can be set
using set_dl_model_layer_param and which ones can be retrieved
using get_dl_model_layer_param or get_dl_layer_param. Note, the
operators set_dl_model_layer_param and get_dl_model_layer_param
require a model created by create_dl_model.
| Layer Parameters | set |
get |
|---|---|---|
'input_layer' (DLLayerInput) |
x |
|
'name' (LayerName) |
x |
x |
| 'axes' | x |
|
| 'starts' | x |
|
| 'ends' | x |
|
| 'steps' | x |
|
'output_layer' (DLLayerSlice) |
x |
|
| 'shape' | x |
|
| 'type' | x |
| Generic Layer Parameters | set |
get |
|---|---|---|
| 'is_inference_output' | x |
x |
| 'num_trainable_params' | x |
Execution information🔗
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🔗
DLLayerInput (input_control) dl_layer → (handle)
Feeding layer.
Default: 'DLLayerInput'
LayerName (input_control) string → (string)
Name of the output layer.
Axes (input_control) number(-array) → (integer / string)
Axes used for slicing.
Default: 0
Suggested values: 0, 'width', 1, 'height', 2, 'depth'
StartIndices (input_control) number(-array) → (integer)
Starting indices of corresponding axes (inclusive).
Default: 0
EndIndices (input_control) number(-array) → (integer)
Ending indices of corresponding axes (exclusive).
Default: 1
Steps (input_control) number(-array) → (integer)
Slicing steps.
Default: []
GenParamName (input_control) attribute.name(-array) → (string)
Generic input parameter names.
Default: []
List of values: 'axes_format', 'is_inference_output'
GenParamValue (input_control) attribute.value(-array) → (string / integer / real)
Generic input parameter values.
Default: []
Suggested values: 'true', 'false', 'nchw', 'whcn'
DLLayerSlice (output_control) dl_layer → (handle)
Output layer.
Module🔗
Deep Learning Professional