Intermediate Representation Notation Reference Catalog

NOTE: This IR Notation Reference is no longer supported since the new concept of operation sets is introduced in OpenVINO 2020.1 version. For a complete list of supported operations, see the Intermediate Representation and Operation Sets topic.

Table of Сontents

Activation Layer

Back to top

Name: Activation

Category: Activation

Short description: Activation layer represents an activation function of each neuron in a layer, which is used to add non-linearity to the computational flow.

Detailed description: Reference

Parameters: Activation layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: type
    • Description: type represents particular activation function. For example, type equal to sigmoid means that the neurons of this layer have a sigmoid activation function.
    • Range of values:
      • sigmoid - sigmoid activation function. Learn more from the Detailed description section.
      • tanh - tanh activation function. Learn more from the Detailed description section.
      • elu - elu activation function. Learn more from the Detailed description section.
      • relu6 - relu6 activation function
      • not - logical NOT function
      • exp - exponent function
    • Type: string
    • Default value: None
    • Required: yes

Mathematical Formulation

  • Sigmoid function:

    \[ f(x) = \frac{1}{1+e^{-x}} \]

  • Tahn function:

    \[ f (x) = \frac{2}{1+e^{-2x}} - 1 = 2sigmoid(2x) - 1 \]

  • Elu function:

    \[ f(x) = \left\{\begin{array}{ll} e^{x} - 1 \quad \mbox{if } x < 0 \\ x \quad \mbox{if } x \geq 0 \end{array}\right. \]

  • Relu6 function:

    \[ f(x) = min(max(0, x), 6) \]

Inputs:

  • 1: Multidimensional input blob. Required.

Example

<layer ... type="Activation" ... >
<data type="sigmoid" />
<input> ... </input>
<output> ... </output>
</layer>

ArgMax Layer

Back to top

Name: ArgMax

Category: Layer

Short description: ArgMax layer computes indexes and values of the top_k maximum values for each datum across all dimensions CxHxW.

Detailed description: ArgMax layer is used after a classification layer to produce a prediction. If the parameter out_max_val is 1, output is a vector of pairs (max_ind, max_val) for each batch. The axis parameter specifies an axis along which to maximize.

Parameters: ArgMax layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: out_max_val
    • Description: If out_max_val is 1, the output is a list of pairs (max_ind, max_val). If out_max_val is 0, the output is a list of indexes of size top_k.
    • Range of values: 0 or 1
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: top_k
    • Description: top_k is the number of elements to save in output.
    • Range of values: a positive integer
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: axis
    • Description: If axis is set, maximizes along the specified axis, else maximizes the flattened trailing dimensions for each index of the first / num dimension.
    • Range of values: an integer. Negative value means counting dimension from the end.
    • Type: int
    • Default value: None
    • Required: no

Inputs:

  • 1: 4D input blob. Required.

Mathematical Formulation

ArgMax generally does the following with the input blobs:

\[ o_{i} = \left\{ x| x \in S \wedge \forall y \in S : f(y) \leq f(x) \right\} \]

Example

<layer ... type="ArgMax" ... >
<data top_k="10" out_max_val="1" axis="-1"/>
<input> ... </input>
<output> ... </output>
</layer>

BatchNormalization Layer

Back to top

Name: BatchNormalization

Category: Normalization

Short description: Reference

Detailed description: Reference

Parameters: BatchNormalization layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: epsilon
    • Description: epsilon is the number to be added to the variance to avoid division by zero when normalizing a value. For example, epsilon equal to 0.001 means that 0.001 is added to the variance.
    • Range of values: a positive floating-point number
    • Type: float
    • Default value: None
    • Required: yes

Inputs:

  • 1: 4D input blob. Required.

Mathematical Formulation

BatchNormalization normalizes the output in each hidden layer.

  • Input: Values of \(x\) over a mini-batch:

    \[ \beta = \{ x_{1...m} \} \]

  • Parameters to learn: \( \gamma, \beta\)
  • Output:

    \[ \{ o_{i} = BN_{\gamma, \beta} ( b_{i} ) \} \]

  • Mini-batch mean:

    \[ \mu_{\beta} \leftarrow \frac{1}{m}\sum_{i=1}^{m}b_{i} \]

  • Mini-batch variance:

    \[ \sigma_{\beta }^{2}\leftarrow \frac{1}{m}\sum_{i=1}^{m} ( b_{i} - \mu_{\beta} )^{2} \]

  • Normalize:

    \[ \hat{b_{i}} \leftarrow \frac{b_{i} - \mu_{\beta}}{\sqrt{\sigma_{\beta }^{2} + \epsilon }} \]

  • Scale and shift:

    \[ o_{i} \leftarrow \gamma\hat{b_{i}} + \beta = BN_{\gamma ,\beta } ( b_{i} ) \]

Example

<layer ... type="BatchNormalization" ... >
<data epsilon="9.99e-06" />
<input> ... </input>
<output> ... </output>
</layer>

BinaryConvolution Layer

Back to top

Name: BinaryConvolution

Category: Layer

Short description: BinaryConvolution convolution with binary weights

Parameters: BinaryConvolution layer parameters are specified in the data node, which is a child of the layer node. The layer has the same parameters as a regular Convolution layer and several unique parameters.

  • Parameter name: input
    • Description: input is the number of input channels.
    • Range of values: a positive integer
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: mode
    • Description: mode defines how input tensor 0/1 values and weights 0/1 are interpreted as real numbers and how the result is computed.
    • Range of values:
      • xnor-popcount
    • Type: string
    • Default value: None
    • Required: yes
  • Parameter name: pad_value
    • Description: pad_value is a floating-point value used to fill pad area.
    • Range of values: a floating-point number
    • Type: float
    • Default value: None
    • Required: yes

Inputs:

  • 1: 4D input blob containing integer or floats; filled with 0/1 values. 0 means -1, 1 means 1 for mode="xnor-popcount". Required.

Bucketize Layer

Back to top

Name: Bucketize

Category: Layer

Short description: Bucketize bucketizes the input based on boundaries. This is an equivalent to np.digitize.

Detailed description: Reference

  • Parameter name: with_right_bound
    • Description: Indicates whether the intervals include the right or the left bucket edge.
    • Range of values: True or False
    • Type: bool
    • Default value: None
    • Required: yes

Inputs:

  • 1: N-D tensor. Input tensor for the bucketization. It contains with float or integer types. Required.
  • 2: 1-D tensor. Sorted boundaries of the buckets. It contains with a float type. Required.

Outputs:

  • 1: Output tensor with bucket indices for each element of the first input tensor. If the second input is empty, the bucket indice for all elements is equal to 0. The output tensor shape is the same as the first input tensor shape.

Clamp Layer

Back to top

Name: Clamp

Category: Layer

Short description: Clamp layer represents clipping activation operation.

Detailed description: Reference

Parameters: Clamp layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: min
    • Description: min is the lower bound of values in the output. Any value in the input that is smaller than the bound is replaced by the min value. For example, min equal to 10.0 means that any value in the input that is smaller than the bound is replaced by 10.0.
    • Range of values: a non-negative floating-point number
    • Type: float
    • Default value: 0.0
    • Required: yes
  • Parameter name: max
    • Description: max is the upper bound of values in the output. Any value in the input that is greater than the bound, is replaced by the max value. For example, max equal to 50.0 means that any value in the input that is greater than the bound is replaced by 50.0.
    • Range of values: a positive floating-point number
    • Type: float
    • Default value: 6.0
    • Required: yes

Inputs:

  • 1: Multidimensional input blob. Required.

Mathematical Formulation

Clamp generally does the following with the input blobs:

\[ out_i=\left\{\begin{array}{ll} max\_value \quad \mbox{if } \quad input_i>max\_value \\ min\_value \quad \mbox{if } \quad input_i \end{array}\right. \]

Example

<layer ... type="Clamp" ... >
<data min="10" max="50" />
<input> ... </input>
<output> ... </output>
</layer>

Broadcast

Back to top

Category: Layer

Short description: Broadcast replicates data on the first input to fit a given shape.

Detailed description:

Broadcast takes the first tensor and, following the NumPy broadcasting rules specification, builds a new tensor with shape matching the second input tensor. The second input value represents desired output shape.

Parameters: Broadcast layer does not have parameters.

Inputs:

  • 1: Source tensor that is being broadcasted. Required.
  • 2: 1D tensor describing output shape. Required.

Outputs:

  • 1: Output tensor with replicated content from the first tensor with shape defined by the second input.

Example

<layer ... type="Broadcast" ...>
<input>
<port id="0">
<dim>16</dim>
<dim>1</dim>
<dim>1</dim>
</port>
<port id="1">
<dim>4</dim> <!--The tensor contains 4 elements: 1, 16, 50, 50 -->
</port>
</input>
<output>
<port id="2">
<dim>1</dim>
<dim>16</dim>
<dim>50</dim>
<dim>50</dim>
</port>
</output>
</layer>

Concat Layer

Back to top

Name: Concat

Category: Layer

Short description: Reference

Parameters: Concat layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: axis
    • Description: axis is the number of axis over which input blobs are concatenated. For example, axis equal to 1 means that input blobs are concatenated over the first axis.
    • Range of values: a non-negative integer
    • Type: int
    • Default value: 1
    • Required: yes

Inputs:

  • 1: Multidimensional input blob. Required.
  • 2: Multidimensional input blob. Required.

Mathematical Formulation

Axis parameter specifies a blob dimension to concatenate values over. For example, for two input blobs B1xC1xH1xW1 and B2xC2xH2xW2, if axis="1", the output blob is B1xC1+C2xH1xW1. This is only possible if B1=B2, H1=H2, W1=W2.

Example

<layer ... type="Concat" ... >
<data axis="1"/>
<input> ... </input>
<output> ... </output>
</layer>

Const Layer

Back to top

Name: Const

Category: Layer

Short description: Const layer produces a blob with a constant value specified in the blobs section.

Parameters: Const layer does not have parameters.

Example

<layer ... type="Const" ...>
<output>
<port id="1">
<dim>3</dim>
<dim>100</dim>
</port>
</output>
<blobs>
<custom offset="..." size="..."/>
</blobs>
</layer>

Convolution Layer

Back to top

Name: Convolution

Category: Layer

Short description: Reference

Detailed description: Reference

Parameters: Convolution layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: strides
    • Description: strides is a distance (in pixels) to slide the filter on the feature map over the (z, y, x) axes for 3D convolutions and (y, x) axes for 2D convolutions. For example, strides equal to "4,2,1" means sliding the filter four pixels at a time over depth dimension, two pixels over height dimension, and one pixel over width dimension.
    • Range of values: a list of non-negative integers
    • Type: int[]
    • Default value: a list of 1 with length equal to the number of convolution kernel dimensions
    • Required: no
  • Parameter name: pads_begin
    • Description: pads_begin is the number of pixels to add to the beginning of each axis. For example, pads_begin equal to "1,2" means adding one pixel to the top of the input and two pixels to the left of the input.
    • Range of values: a list of non-negative integers
    • Type: int[]
    • Default value: a list of 0 with length equal to the number of convolution kernel dimensions
    • Required: no
  • Parameter name: pads_end
    • Description: pads_end is the number of pixels to add to the end of each axis. For example, pads_end equal to "1,2" means adding one pixel to the bottom of the input and two pixels to the right of the input.
    • Range of values: a list of non-negative integers
    • Type: int[]
    • Default value: a list of 0 with length equal to the number of convolution kernel dimensions
    • Required: no
  • Parameter name: kernel
    • Description: kernel is a size of each filter. For example, kernel equal to "2,3" means that each filter has height equal to 2 and width equal to 3.
    • Range of values: a list of non-negative integers
    • Type: int[]
    • Default value: None
    • Required: yes
  • Parameter name: output
    • Description: output is a number of output feature maps in the output. If group parameter value is greater than 1, output still matches the number of output features regardless of group value. For example, output equal to 1 means that there is one output feature map in a layer.
    • Range of values: a non-negative integer
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: group
    • Description: group is the number of groups which output and input should be split into. For example, group equal to 1 means that all filters are applied to the whole input (usual convolution), group equal to 2 means that both input and output channels are separated into two groups and the i-th output group is connected to the i-th input group channel. group equal to a number of output feature maps implies depth-wise separable convolution. For more information, see the Reference.
    • Range of values: an integer
    • Type: int
    • Default value: 1
    • Required: no
  • Parameter name: dilations
    • Description: dilations is a distance in width and height between elements (weights) in the filter. For example, dilations equal to "1,1" means that all elements in the filter are neighbors, so it is the same as the usual convolution. dilations equal to "2,2" means that all elements in the filter are matched to the elements in the input matrix separated by one pixel.
    • Range of values: a non-negative integer
    • Type: int[]
    • Default value: a list of 1 with length equal to the number of convolution kernel dimensions
    • Required: no
  • Parameter name: auto_pad
    • Description: auto_pad defines how the padding is calculated. Possible values:
      • Not specified: use explicit padding values
      • same_upper/same_lower: add paddings to the input to match the output size. In case of odd padding value, an extra padding is added to the beginning if auto_pad="same_upper" or to the end if auto_pad="same_lower".
      • valid: do not use padding
    • Type: string
    • Default value: None
    • Required: no

Inputs:

  • 1: 4D or 5D input blob. Required.

Weights Layout

Weights layout is GOIYX (GOIZYX for 3D convolution), which means that X changes the fastest, then Y, Input and Output, Group.

Mathematical Formulation

  • For the convolutional layer, the number of output features in each dimension is calculated as:

    \[ n_{out} = \left ( \frac{n_{in} + 2p - k}{s} \right ) + 1 \]

  • The receptive field in each layer is calculated as:
    • Jump in the output feature map:

      \[ j_{out} = j_{in} * s \]

    • Size of the receptive field of output feature:

      \[ r_{out} = r_{in} + ( k - 1 ) * j_{in} \]

    • Center position of the receptive field of the first output feature:

      \[ start_{out} = start_{in} + ( \frac{k - 1}{2} - p ) * j_{in} \]

    • Output is calculated as:

      \[ out = \sum_{i = 0}^{n}w_{i}x_{i} + b \]

Example

<layer ... type="Convolution" ... >
<data auto_pad="same_upper" dilations="1,1" group="3" kernel="7,7" output="24" pads_begin="2,2" pads_end="3,3" strides="2,2"/>
<input> ... </input>
<output> ... </output>
<weights ... />
<biases ... />
</layer>

Crop (Type 1) Layer

Back to top

Name: Crop

Category: Layer

Short description: Crop layer changes selected dimensions of the input blob according to the specified parameters.

Parameters: Crop layer parameters are specified in the data section, which is a child of the layer node. Crop Type 1 layer takes two input blobs, and the shape of the second blob specifies the Crop size. The Crop layer of this type supports shape inference.

  • Parameter name: axis
    • Description: axis is the number of a dimension to crop. For example, axis equal to [1] means that the first dimension is cropped.
    • Range of values: a list of unique integers, where each element is greater than or equal to 0 and less than input shape length
    • Type: int[]
    • Default value: [1]
    • Required: yes
  • Parameter name: offset
    • Description: offset is the starting point for crop in the input blob. For example, offset equal to 2 means that crop starts from the second value of a specified axis.
    • Range of values: a list of integers of the length equal to the length of the axis attribute. In the list, offset[i] is greater than or equal to 0 and less than or equal to input_shape[axis[i]] - crop_size[axis[i]], where crop_size is the shape of the second input.
    • Type: int[]
    • Default value: None
    • Required: yes

Inputs

  • 1: Multidimensional input blob
  • 2: Shape of this input will be used for crop

Example

<layer id="39" name="score_pool4c" precision="FP32" type="Crop">
<data axis="2,3" offset="0,0"/>
<input>
<port id="0">
<dim>1</dim>
<dim>21</dim>
<dim>44</dim>
<dim>44</dim>
</port>
<port id="1">
<dim>1</dim>
<dim>21</dim>
<dim>34</dim>
<dim>34</dim>
</port>
</input>
<output>
<port id="2">
<dim>1</dim>
<dim>21</dim>
<dim>34</dim>
<dim>34</dim>
</port>
</output>
</layer>

Crop (Type 2) Layer

Back to top

Name: Crop

Category: Layer

Short description: Crop layer changes selected dimensions of the input blob according to the specified parameters.

Parameters: Specify parameters for the Crop layer in the data section, which is a child of the layer node. Crop Type 2 layer takes one input blob to crop. The Crop layer of this type supports shape inference only when shape propagation is applied to dimensions not specified in the axis attribute.

  • Parameter name: axis
    • Description: axis is the number of a dimension to crop. For example, axis equal to [1] means that the first dimension is cropped.
    • Range of values: a list of unique integers, where each element is greater than or equal to 0 and less than input shape length
    • Type: int[]
    • Default value: [1]
    • Required: yes
  • Parameter name: offset
    • Description: offset is the starting point for crop in the input blob. For example, offset equal to 2 means that cropping starts from the second value of the specified axis.
    • Range of values: a list of integers with the length equal to the length of axis attribute, where offset[i] is greater than or equal to 0 and less or equal to input_shape[axis[i]] - dim[i]
    • Type: int[]
    • Default value: None
    • Required: yes
  • Parameter name: dim
    • Description: dim is the resulting size of the output blob for the specified axis. For example, dim equal to [88] means that the output blob gets the dimension equal to 88 for the specified axis.
    • Range of values: a list of integers
    • Type: int[]
    • Default value: None
    • Required: yes

Example

<layer id="39" name="score_pool4c" precision="FP32" type="Crop">
<data axis="2,3" offset="0,0" dim="34,34"/>
<input>
<port id="0">
<dim>1</dim>
<dim>21</dim>
<dim>44</dim>
<dim>44</dim>
</port>
</input>
<output>
<port id="1">
<dim>1</dim>
<dim>21</dim>
<dim>34</dim>
<dim>34</dim>
</port>
</output>
</layer>

Crop (Type 3) Layer

Back to top

Name: Crop

Category: Layer

Short description: Crop layer changes selected dimensions of the input blob according to the specified parameters.

Parameters: Crop layer parameters are specified in the data section, which is a child of the layer node. Crop Type 3 layer takes one input blob to crop. The Crop layer of this type supports shape inference.

  • Parameter name: axis
    • Description: axis is the number of a dimension to crop. For example, axis equal to [1] means that the first dimension is cropped.
    • Range of values: a list of unique integers, where each element is greater than or equal to 0 and less than input shape length
    • Type: int[]
    • Default value: [1]
    • Required: yes
  • Parameter name: crop_begin
    • Description: crop_begin specifies the starting offset for crop in the input blob for a specified axes.
    • Range of values: a list of integers, where crop_begin[i] is greater than or equal to 0 and less than input_shape[axis[i]] - crop_end[i]
    • Type: int[]
    • Default value: None
    • Required: yes
  • Parameter name: crop_end
    • Description: crop_end specifies the ending offset for crop in the input blob for the specified axes.
    • Range of values: a list of integers, where crop_end[i] is greater than or equal to 0 and less than input_shape[axis[i]] - crop_begin[i]
    • Type: int[]
    • Default value: None
    • Required: yes

Example

<layer id="39" name="score_pool4c" precision="FP32" type="Crop">
<data axis="2,3" crop_begin="4,4" crop_end="6,6"/>
<input>
<port id="0">
<dim>1</dim>
<dim>21</dim>
<dim>44</dim>
<dim>44</dim>
</port>
</input>
<output>
<port id="1">
<dim>1</dim>
<dim>21</dim>
<dim>34</dim>
<dim>34</dim>
</port>
</output>
</layer>

CTCGreedyDecoder Layer

Back to top

Name: CTCGreedyDecoder

Category: Layer

Short description: CTCGreedyDecoder performs greedy decoding on the logits given in input (best path).

Detailed description: Reference

Parameters: CTCGreedyDecoder layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: ctc_merge_repeated
    • Description: ctc_merge_repeated is a flag for merging repeated labels during the CTC calculation.
    • Range of values: 0 or 1
    • Type: int
    • Default value: None
    • Required: yes

Mathematical Formulation

Given an input sequence \(X\) of length \(T\), CTCGreadyDecoder assumes the probability of a length \(T\) character sequence \(C\) is given by

\[ p(C|X) = \prod_{t=1}^{T} p(c_{t}|X) \]

Example

<layer ... type="CTCGreadyDecoder" ... >
<data stride="1"/>
<input> ... </input>
<output> ... </output>
</layer>

Deconvolution Layer

Back to top

Name: Deconvolution

Category: Layer

Short description: Deconvolution layer is applied for upsampling the output to the higher image resolution.

Detailed description: Reference

Parameters: Deconvolution layer parameters should be specified in the data node, which is a child of the layer node.

  • Parameter name: strides
    • Description: strides is a distance (in pixels) to slide the filter on the feature map over the (z, y, x) axes for 3D deconvolutions and (y, x) axes for 2D deconvolutions. For example, strides equal to "4,2,1" means sliding the filter four pixels at a time over depth dimension, two pixels over height dimension, and one pixel over width dimension.
    • Range of values: a list of non-negative integers
    • Type: int[]
    • Default value: a list of 1 with length equal to the number of convolution kernel dimensions
    • Required: no
  • Parameter name: pads_begin
    • Description: pads_begin is the number of pixels to add to the beginning of each axis. For example, pads_begin equal to "1,2" means adding one pixel to the top of the input and two pixels to the left of the input.
    • Range of values: a list of non-negative integers
    • Type: int[]
    • Default value: a list of 0 with length equal to the number of convolution kernel dimensions
    • Required: no
  • Parameter name: pads_end
    • Description: pads_end is the number of pixels to add to the end of each axis. For example, pads_end equal to "1,2" means adding one pixel to the bottom of the input and two pixels to the right of the input.
    • Range of values: a list of non-negative integers
    • Type: int[]
    • Default value: a list of 1 with length equal to the number of convolution kernel dimensions
    • Required: no
  • Parameter name: kernel
    • Description: kernel is a size of each filter. For example, kernel equal to "2,3" means that each filter has height equal to 2 and width equal to 3.
    • Range of values: a list of positive integers
    • Type: int[]
    • Default value: None
    • Required: yes
  • Parameter name: output
    • Description: output is the number of output feature maps in the output. If group parameter value is greater than 1, output still matches the number of output features regardless of the group value. For example, output equal to 1 means that there is one output feature map in a layer.
    • Range of values: a non-negative integer
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: group
    • Description: group denotes the number of groups to which output and input should be split. For example, group equal to 1 means that all filters are applied to the whole input (usual convolution), group equal to 2 means that both input and output channels are separated into 2 groups and i-th output group is connected to i-th input group channels. group equal to a number of output feature maps implies depth-wise separable convolution. For more information, see the Reference.
    • Range of values: a non-negative integer
    • Type: int
    • Default value: 1
    • Required: no
  • Parameter name: dilations
    • Description: dilations is the distance in width and height between elements (weights) in the filter. For example, dilation equal to "1,1" means that all elements in the filter are neighbors, so it is the same as the usual convolution. dilation equal to "2,2" means that all elements in the filter are matched to the elements in the input matrix separated by one pixel.
    • Range of values: a list of non-negative integers
    • Type: int[]
    • Default value: a list of 1 with length equal to the number of convolution kernel dimensions
    • Required: no
  • Parameter name: auto_pad
    • Description: auto_pad defines how the padding is calculated.
    • Range of values:
      • Not specified: use explicit padding values.
      • same_upper/same_lower: add paddings to the input to match the output size. In case of odd padding value, an extra padding is added to the beginning if auto_pad="same_upper" or to the end if auto_pad="same_lower".
      • valid: do not use padding
    • Type: string
    • Default value: None
    • Required: no

Inputs:

  • 1: 4D or 5D blob with input data. Required.

Weights Layout

Weights layout is GOIYX, which means that X changes the fastest, then Y, Input and Output, Group.

Mathematical Formulation

Deconvolution is also called transpose convolution and performs operation that is reverse to convolution. The number of output features for each dimensions is calculated as:

\[S_{o}=stride(S_{i} - 1 ) + S_{f} - 2pad \]

Where \(S\) is the size of output, input, and filter. Output is calculated in the same way as for convolution layer:

\[out = \sum_{i = 0}^{n}w_{i}x_{i} + b\]

Example

<layer ... type="Deconvolution" ...>
<data auto_pad="valid" kernel="2,2,2" output="512" pads_begin="0,0,0" pads_end="0,0,0" strides="2,2,2"/>
<input>
<port id="0">
<dim>1</dim>
<dim>512</dim>
<dim>8</dim>
<dim>8</dim>
<dim>8</dim>
</port>
</input>
<output>
<port id="3">
<dim>1</dim>
<dim>512</dim>
<dim>16</dim>
<dim>16</dim>
<dim>16</dim>
</port>
</output>
<blobs>
<weights offset="..." size="..."/>
<biases offset="..." size="..."/>
</blobs>
</layer>

DeformableConvolution Layer

Back to top

Name: DeformableConvolution

Category: Layer

Short description: DeformableConvolution convolution layer enhances the transformation modeling capacity of CNNs.

Detailed description: Reference

Parameters: DeformableConvolution layer parameters are specified in the data node, which is a child of the layer node. The layer has the same parameters as a regular Convolution layer and several unique parameters.

  • Parameter name: num_deformable_group
    • Description: num_deformable_group is the number of deformable group partitions.
    • Range of values: a non-negative integer
    • Type: int
    • Default value: 1
    • Required: no

Inputs:

  • 1: 4D or 5D blob with input data. Required.
  • 2: Input offset to the DeformableConvolution

Weights Layout

Weights layout is GOIYX (GOIZYX for 3D convolution), which means that X changes the fastest, then Y, Input and Output, Group.

Example

<layer ... type="DeformableConvolution">
<data deformable_group="4" dilations="2,2" group="1" kernel="3,3" output="512" pads_begin="2,2" pads_end="2,2" strides="1,1"/>
<input>
<port id="0">
<dim>1</dim>
<dim>512</dim>
<dim>40</dim>
<dim>27</dim>
</port>
<port id="1">
<dim>1</dim>
<dim>72</dim>
<dim>40</dim>
<dim>27</dim>
</port>
</input>
<output>
<port id="3">
<dim>1</dim>
<dim>512</dim>
<dim>40</dim>
<dim>27</dim>
</port>
</output>
<blobs>
<weights offset="121799456" size="9437184"/>
</blobs>
</layer>

DepthToSpace Layer

Back to top

Name: DepthToSpace

Category: Layer

Short description: DepthToSpace layer rearranges data from the depth dimension of the input blob into spatial dimensions.

Detailed description: DepthToSpace layer outputs a copy of the input blob, where values from the depth dimension (features) are moved to spatial blocks. Refer to the ONNX* specification for an example of the 4D input blob case.

Parameters: DepthToSpace layer parameters are specified parameters in the data node, which is a child of the layer node.

  • Parameter name: block_size
    • Description: block_size specifies the size of the value block to be moved. The depth dimension size must be evenly divided by block_size ^ (len(input.shape) - 2).
    • Range of values: a positive integer
    • Type: int
    • Default value: 1
    • Required: no

Inputs:

  • 1: 3D+ blob with input data. Required.

Mathematical Formulation

The operation is equivalent to the following transformation of the input blob x with K spatial dimensions of shape [N, C, D1, D2, D3 , ... , DK]:

x' = reshape(x, [N, block_size, block_size, ... , block_size, D1 * block_size, D2 * block_size, ... Dk * block_size])
x'' = transpose(x', [0, K + 1, K + 2, 1, K + 3, 2, K + 4, 3, ... K + K + 1, K])
y = reshape(x'', [N, C / block_size ^ K, D1 * block_size, D2 * block_size, D3 * block_size, ... , DK * block_size])

Example

<layer ... type="DepthToSpace">
<data block_size="2"/>
<input>
<port id="0">
<dim>5</dim>
<dim>4</dim>
<dim>2</dim>
<dim>3</dim>
</port>
</input>
<output>
<port id="1">
<dim>5</dim> <!-- input.shape[0] -->
<dim>1</dim> <!-- input.shape[1] / (block_size ^ 2) -->
<dim>4</dim> <!-- input.shape[2] * block_size -->
<dim>6</dim> <!-- input.shape[3] * block_size -->
</port>
</output>
</layer>

DetectionOutput Layer

Back to top

Name: DetectionOutput

Category: Layer

Short description: DetectionOutput layer performs non-maximum suppression to generate the detection output using information on location and confidence predictions.

Detailed description: Reference. The layer has three required inputs: blob with box logits, blob with confidence predictions, and blob with box coordinates (proposals). It can have two additional inputs with additional confidence predictions and box coordinates described in the article. The five input version of the layer is supported with MYRIAD plugin only. The output blob contains information about filtered detections described with seven element tuples: [batch_id, class_id, confidence, x_1, y_1, x_2, y_2]. The first tuple with batch_id equal to -1 means end of output.

Parameters: DetectionOutput layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: num_classes
    • Description: num_classes is the number of classes to be predicted.
    • Range of values: a positive integer
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: background_label_id
    • Description: background_label_id is the background label ID. If there is no background class, set it to -1.
    • Range of values: an integer
    • Type: int
    • Default value: 0
    • Required: no
  • Parameter name: top_k
    • Description: top_k is the maximum number of results to keep per batch after NMS step. -1 means keeping all bounding boxes.
    • Range of values: an integer
    • Type: int
    • Default value: -1
    • Required: no
  • Parameter name: variance_encoded_in_target
    • Description: variance_encoded_in_target is a flag that specifies if variance is encoded in target. If flag is 0 (that is, false), you need to adjust the predicted offset accordingly.
    • Range of values: 0 or 1
    • Type: int
    • Default value: 0
    • Required: no
  • Parameter name: keep_top_k
    • Description: keep_top_k is the maximum number of bounding boxes per batch to keep after NMS step. -1 means keeping all bounding boxes.
    • Range of values: an integer
    • Type: int
    • Default value: -1
    • Required: yes
  • Parameter name: code_type
    • Description: code_type is a coding method for bounding boxes.
    • Range of values: "caffe.PriorBoxParameter.CENTER_SIZE", "caffe.PriorBoxParameter.CORNER"
    • Type: string
    • Default value: caffe.PriorBoxParameter.CORNER
    • Required: no
  • Parameter name: share_location
    • Description: share_location is a flag that specifies if bounding boxes are shared among different classes.
    • Range of values: 0 or 1
    • Type: int
    • Default value: 1
    • Required: no
  • Parameter name: nms_threshold
    • Description: nms_threshold is the threshold to be used in the NMS stage.
    • Range of values: a floating-point number
    • Type: float
    • Default value: None
    • Required: yes
  • Parameter name: confidence_threshold
    • Description: confidence_threshold is a threshold to filter out detections with smaller confidence. If not set, all boxes are used.
    • Range of values: a floating-point number
    • Type: float
    • Default value: -FLT_MAX
    • Required: no
  • Parameter name: clip_after_nms
    • Description: clip_after_nms is a flag that specifies whether to perform clip bounding boxes after non-maximum suppression or not.
    • Range of values: 0 or 1
    • Type: int
    • Default value: 0
    • Required: no
  • Parameter name: clip_before_nms
    • Description: clip_before_nms is a flag that specifies whether to clip bounding boxes before non-maximum suppression or not.
    • Range of values: 0 or 1
    • Type: int
    • Default value: 0
    • Required: no
  • Parameter name: decrease_label_id
    • Description: decrease_label_id is a flag that denotes how to perform NMS.
    • Range of values:
      • 0 - perform NMS like in Caffe*
      • 1 - perform NMS like in MxNet*
    • Type: int
    • Default value: 0
    • Required: no
  • Parameter name: normalized
    • Description: normalized is a flag that specifies whether input blobs with boxes are normalized. If blobs are not normalized, the input_height and input_width parameters are used to normalize box coordinates.
    • Range of values: 0 or 1
    • Type: int
    • Default value: 0
    • Required: no
  • Parameter name: input_height
    • Description: input_height is the height of an input image. If the normalized is 1, input_height is ignored.
    • Range of values: a positive integer
    • Type: int
    • Default value: 1
    • Required: no
  • Parameter name: input_width
    • Description: input_width is the width of an input image. If the normalized is 1, input_width is ignored.
    • Range of values: a positive integer
    • Type: int
    • Default value: 1
    • Required: no
  • Parameter name: objectness_score
    • Description: objectness_score is the threshold to sort out confidence predictions. Used only when the DetectionOutput layer has five inputs.
    • Range of values: a non-negative floating-point number
    • Type: float
    • Default value: 0.0
    • Required: no

Inputs:

  • 1: 2D input blob with box logits. Required.
  • 2: 2D input blob with class predictions. Required.
  • 3: 3D input blob with proposals. Required.
  • 4: 2D input blob with additional class predictions information described in the article. Optional.
  • 5: 2D input blob with additional box predictions information described in the article. Optional.

Mathematical Formulation

At each feature map cell, DetectionOutput predicts the offsets relative to the default box shapes in the cell, as well as the per-class scores that indicate the presence of a class instance in each of those boxes. Specifically, for each box out of k at a given location, DetectionOutput computes class scores and the four offsets relative to the original default box shape. This results are a total of \((c + 4)k\) filters that are applied around each location in the feature map, yielding \((c + 4)kmn\) outputs for a m * n feature map.

Example

<layer ... type="DetectionOutput" ... >
<data num_classes="21" share_location="1" background_label_id="0" nms_threshold="0.450000" top_k="400" input_height="1" input_width="1" code_type="caffe.PriorBoxParameter.CENTER_SIZE" variance_encoded_in_target="0" keep_top_k="200" confidence_threshold="0.010000"/>
<input> ... </input>
<output> ... </output>
</layer>

Erf Layer

Back to top

Name: Erf

Category: Layer

Short description: Erf layer computes the Gauss error function of input element-wise.

Detailed Description: Reference

Parameters: Erf layer does not have parameters.

Inputs:

  • 1: Input tensor X of any floating-point type. Required.

Outputs:

  • 1: Result of Erf function applied on input tensor x. Floating point tensor with shape and type matching input tensor. Required.

Mathematical Formulation

For each element from an input tensor, Erf layer calculates corresponding element in the output tensor by the formula:

\[ erf(x) = \frac{2}{\sqrt{\pi}} \int_0^x e^{-t^2} dt \]

Example

<layer ... type="Erf" ... >
<input>
<port id="0">
<dim>5</dim>
<dim>4</dim>
</port>
</input>
<output>
<port id="1">
<dim>5</dim>
<dim>4</dim>
</port>
</output>
</layer>

Eltwise Layer

Back to top

Name: Eltwise

Category: Layer

Short description: Eltwise layer performs element-wise operation specified in parameters, over given inputs.

Parameters: Eltwise layer parameters are specified in the data node, which is a child of the layer node. Eltwise accepts two inputs of arbitrary number of dimensions. The operation supports broadcasting input blobs according to the NumPy specification.

  • Parameter name: operation
    • Description: operation is a mathematical operation to be performed over inputs.
    • Range of values:
      • sum - summation
      • sub - subtraction
      • mul - multiplication
      • div - division
      • max - maximum
      • min - minimum
      • squared_diff - squared difference
      • floor_mod - reminder of division
      • pow - power
      • logical_and - logical AND
      • logical_or - logical OR
      • logical_xor - logical XOR
      • less - less
      • less_equal - less or equal
      • greater - greater
      • greater_equal - greater equal
      • equal - equal
      • not_equal - not equal
    • Type: string
    • Default value: sum
    • Required: no

Inputs

  • 1: Multidimensional input blob. Required.
  • 2: Multidimensional input blob. Required.

Mathematical Formulation Eltwise does the following with the input blobs:

\[ o_{i} = f(b_{i}^{1}, b_{i}^{2}) \]

where \(b_{i}^{1}\) - first blob \(i\)-th element, \(b_{i}^{2}\) - second blob \(i\)-th element, \(o_{i}\) - output blob \(i\)-th element, \(f(a, b)\) - is a function that performs an operation over its two arguments \(a, b\).

Example

<layer ... type="Eltwise" ... >
<data operation="sum"/>
<input> ... </input>
<output> ... </output>
</layer>

Fill Layer

Back to top

Name: Fill

Category: Layer

Short description: Fill layer generates a blob of the specified shape filled with the specified value.

Parameters: Fill layer does not have parameters.

Inputs:

  • 1: 1D blob with an output blob shape. Required.
  • 2: 0D blob (constant) with the value for fill. Required.

Example

<layer ... type="Fill">
<input>
<port id="0">
<dim>2</dim> <!-- value: [3,4]-->
</port>
<port id="1"/> <!-- value 5-->
</input>
<output>
<port id="2"> <!-- output value is: [[5,5,5,5],[5,5,5,5],[5,5,5,5]]-->
<dim>3</dim>
<dim>4</dim>
</port>
</output>
</layer>

Flatten Layer

Back to top

Name: Flatten

Category: Layer

Short description: Flatten layer performs flattening of specific dimensions of the input blob.

Parameters: Flatten layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: axis
    • Description: axis specifies the first axis to flatten.
    • Range of values: a non-negative integer
    • Type: int
    • Default value: 0
    • Required: no
  • Parameter name: end_axis
    • Description: end_axis specifies the last dimension to flatten. The value can be negative meaning counting axes from the end.
    • Range of values: an integer
    • Type: int
    • Default value: -1
    • Required: no

Inputs

  • 1: Multidimensional input blob. Required.

Example

<layer ... type="Flatten" ...>
<data axis="1" end_axis="-1"/>
<input>
<port id="0">
<dim>7</dim>
<dim>19</dim>
<dim>19</dim>
<dim>12</dim>
</port>
</input>
<output>
<port id="1">
<dim>7</dim>
<dim>4332</dim>
</port>
</output>
</layer>

FullyConnected Layer

Back to top

Name: FullyConnected

Category: Layer

Short description: Reference

Detailed description: Reference

Parameters: FullyConnected layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: out-size
    • Description: out-size is the length of the output vector. For example, out-size equal to 4096 means that the output vector length is 4096.
    • Range of values: a non-negative integer
    • Type: int
    • Default value: None
    • Required: yes

Inputs

  • 1: 2D or 4D input blob. Required.

Weights Layout

OI, which means that Input changes the fastest, then Output.

Mathematical Formulation

  • If previous layer is FullyConnected:

    \[ y_{i} = f( z_{i} ) \quad with \quad z_{i} = \sum_{j=1}^{m_{1}^{( l-1 )}}w_{i,j}^{( l )}y_{i}^{ ( l -1 )} \]

  • Otherwise:

    \[ y_{i} = f( z_{i} ) \quad with \quad z_{i}^{ ( l )} = \sum_{j=1}^{m_{1}^{( l-1 )}}\sum_{r=1}^{m_{2}^{ ( l-1 )}}\sum_{s=1}^{m_{3}^{ ( l-1 )}}w_{i,j,r,s}^{ ( l )} ( Y_{i}^{ (l-1) })_{r,s} \]

Example

<layer ... type="FullyConnected" ... >
<data out-size="4096"/>
<input> ... </input>
<output> ... </output>
</layer>

Gather Layer

Back to top

Name: Gather

Category: Layer

Short description: Gather layer takes slices of data in the second input blob according to the indexes specified in the first input blob. The output blob shape is input2.shape[:axis] + input1.shape + input2.shape[axis + 1:].

Parameters: Gather layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: axis
    • Description: axis is a dimension index to gather data from. For example, axis equal to 1 means that gathering is performed over the first dimension.
    • Range of values: an integer in the range [-len(input2.shape), len(input2.shape) - 1].
    • Type: int
    • Default value: None
    • Required: yes

Mathematical Formulation

\[ output[:, ... ,:, i, ... , j,:, ... ,:] = input2[:, ... ,:, input1[i, ... ,j],:, ... ,:] \]

Inputs

  • 1: Multidimensional input blob with indexes to gather. The values for indexes are in the range [0, input1[axis] - 1].
  • 2: Multidimensional input blob with arbitrary data.

Example

<layer id="1" name="gather_node" precision="FP32" type="Gather">
<data axis=1 />
<input>
<port id="0">
<dim>15</dim>
<dim>4</dim>
<dim>20</dim>
<dim>28</dim>
</port>
<port id="1">
<dim>6</dim>
<dim>12</dim>
<dim>10</dim>
<dim>24</dim>
</port>
</input>
<output>
<port id="2">
<dim>6</dim>
<dim>15</dim>
<dim>4</dim>
<dim>20</dim>
<dim>28</dim>
<dim>10</dim>
<dim>24</dim>
</port>
</output>
</layer>

GRN Layer

Back to top

Name: GRN

Category: Normalization

Short description: GRN is the Global Response Normalization with L2 norm (across channels only).

Parameters: GRN layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: bias
    • Description: bias is added to the variance.
    • Range of values: a floating-point number
    • Type: float
    • Default value: None
    • Required: yes

Inputs

  • 1: 2D, 3D or 4D input blob. Required.

Mathematical Formulation

GRN computes the L2 norm by channels for input blob. GRN generally does the following with the input blob:

\[ output_{i} = \frac{input_{i}}{\sqrt{\sum_{i}^{C} input_{i}}} \]

Example

<layer ... type="GRN" ... >
<data bias="1.0"/>
<input> ... </input>
<output> ... </output>
</layer>

GRUCell Layer

Back to top

Name: GRUCell

Category: Layer

Short description: GRUCell layer computes the output using the formula described in the paper.

Parameters: GRUCell layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: hidden_size
    • Description: hidden_size specifies hidden state size.
    • Range of values: a positive integer
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: activations
    • Description: activations specifies activation functions for gates.
    • Range of values: any combination of relu, sigmoid, tanh
    • Type: a list of strings
    • Default value: sigmoid,tanh
    • Required: no
  • Parameter name: activations_alpha, activations_beta
    • Description: activations_alpha, activations_beta parameters of functions
    • Range of values: a list of floating-point numbers
    • Type: float[]
    • Default value: None
    • Required: no
  • Parameter name: clip
    • Description: clip specifies bound values [-C, C] for tensor clipping. Clipping is performed before activations.
    • Range of values: a positive floating-point number
    • Type: float
    • Default value: None
    • Required: no
  • Parameter name: linear_before_reset
    • Description: linear_before_reset flag denotes if the layer behaves according to the modification of GRUCell described in the formula in the ONNX documentation.
    • Range of values: 0 or 1
    • Type: int
    • Default value: 0
    • Required: no

Inputs

  • 1: X - 2D ([batch_size, input_size]) input data. Required.
  • 2: Hi - 2D ([batch_size, hidden_size]) input hidden state data. Required.

Outputs

  • 1: Ho - 2D ([batch_size, hidden_size]) output hidden state.

Example

<layer type="GRUCell">
<data hidden_size="128" linear_before_reset="1"/>
<input>
<port id="0">
<dim>1</dim>
<dim>16</dim>
</port>
<port id="1">
<dim>1</dim>
<dim>128</dim>
</port>
</input>
<output>
<port id="4">
<dim>1</dim>
<dim>128</dim>
</port>
</output>
<blobs>
<weights offset="…" size="…"/>
<biases offset="…" size="…"/>
</blobs>
</layer>

Input Layer

Back to top

Name: Input

Category: Layer

Short description: Input layer specifies input to the model.

Parameters: Input layer does not have parameters.

Example

<layer ... type="Input" ...>
<output>
<port id="0">
<dim>1</dim>
<dim>3</dim>
<dim>224</dim>
<dim>224</dim>
</port>
</output>
</layer>

Interp Layer

Back to top

Name: Interp

Category: Layer

Short description: Interp layer performs bilinear interpolation of the input blob by the specified parameters.

Parameters: Interp layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: height
    • Description: height specifies output height. If the parameter is not set, other parameters are used for output size calculation.
    • Range of values: a positive integer
    • Type: int
    • Default value: None
    • Required: no
  • Parameter name: width
    • Description: width specifies output width. If the parameter is not set, other parameters are used for output size calculation.
    • Range of values: a positive integer
    • Type: int
    • Default value: None
    • Required: no
  • Parameter name: align_corners
    • Description: align_corners is a flag that specifies whether to align corners or not.
    • Range of values: 0 or 1
    • Type: int
    • Default value: 1
    • Required: no
  • Parameter name: pad_beg
    • Description: pad_beg specify the number of pixels to add to the beginning of the image being interpolated.
    • Range of values: a non-negative integer number
    • Type: int
    • Default value: 0
    • Required: yes
  • Parameter name: pad_end
    • Description: pad_end specify the number of pixels to add to the end of the image being interpolated.
    • Range of values: a non-negative integer
    • Type: int
    • Default value: 0
    • Required: yes

Inputs

  • 1: 4D input blob. Required.

Example

<layer ... type="Interp" ...>
<data align_corners="0" pad_beg="0" pad_end="0"/>
<input>
<port id="0">
<dim>1</dim>
<dim>2</dim>
<dim>48</dim>
<dim>80</dim>
</port>
</input>
<output>
<port id="1">
<dim>1</dim>
<dim>2</dim>
<dim>96</dim>
<dim>160</dim>
</port>
</output>
</layer>

LSTMCell Layer

Back to top

Name: LSTMCell

Category: Layer

Short description: LSTMCell layer computes the output using the formula described in the original paper Long Short-Term Memory.

Parameters: LSTMCell layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: hidden_size
    • Description: hidden_size specifies hidden state size.
    • Range of values: a positive integer
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: activations
    • Description: activations specifies activation functions for gates.
    • Range of values: any combination of relu, sigmoid, tanh
    • Type: a list of strings
    • Default value: sigmoid,tanh,tanh
    • Required: no
  • Parameter name: activations_alpha, activations_beta
    • Description: activations_alpha, activations_beta parameters of functions.
    • Range of values: a list of floating-point numbers
    • Type: float[]
    • Default value: None
    • Required: no
  • Parameter name: clip
    • Description: clip specifies bound values [-C, C] for tensor clipping. Clipping is performed before activations.
    • Range of values: a positive floating-point number
    • Type: float
    • Default value: None
    • Required: no

Inputs

  • 1: X - 2D ([batch_size, input_size]) input data. Required.
  • 2: Hi - 2D ([batch_size, hidden_size]) input hidden state data. Required.
  • 3: Ci - 2D ([batch_size, hidden_size]) input cell state data. Required.

Outputs

  • 1: Ho - 2D ([batch_size, hidden_size]) output hidden state.
  • 2: Co - 2D ([batch_size, hidden_size]) output cell state.

Mathematical Formulation

Formula:
* - matrix mult
(.) - eltwise mult
[,] - concatenation
sigm - 1/(1 + e^{-x})
tanh - (e^{2x} - 1)/(e^{2x} + 1)
f = sigm(Wf*[Hi, X] + Bf)
i = sigm(Wi*[Hi, X] + Bi)
c = tanh(Wc*[Hi, X] + Bc)
o = sigm(Wo*[Hi, X] + Bo)
Co = f (.) Ci + i (.) c
Ho = o (.) tanh(Co)

Example

<layer ... type="LSTMCell" ... >
<input> ... </input>
<output> ... </output>
</layer>

Memory Layer

Back to top

Name: Memory

Category: Layer

Short description: Memory layer represents the delay layer in terms of LSTM terminology. For more information about LSTM topologies, please refer to this article.

Detailed description: Memory layer saves the state between two infer requests. In the topology, it is the single layer, however, in the Intermediate Representation, it is always represented as a pair of Memory layers. One of these layers does not have outputs and another does not have inputs (in terms of the Intermediate Representation).

Parameters: Memory layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: id
    • Description: id is the ID of the pair of Memory layers. Two layers with the same value of the id parameter are paired.
    • Range of values: any combination of Latin characters, numbers, and underscores (_) in the string format
    • Type: string
    • Default value: None
    • Required: yes
  • Parameter name: index
    • Description: index specifies whether the given layer is input or output. For example, index equal to 0 means the layer is output.
    • Range of values:
      • 0 - current layer is output
      • 1 - current layer is input
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: size
    • Description: size is the size of the group. For example, size equal to 2 means this group is a pair.
    • Range of values: only 2 is supported
    • Type: int
    • Default value: None
    • Required: yes

Mathematical Formulation

Memory saves data from the input blob.

Example

<layer ... type="Memory" ... >
<data id="r_27-28" index="0" size="2" />
<input> ... </input>
<output> ... </output>
</layer>

MVN Layer

Back to top

Name: MVN

Category: Normalization

Short description: Reference

Parameters: MVN layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: across_channels
    • Description: across_channels is a flag that specifies whether mean values are shared across channels. For example, across_channels equal to 0 means that mean values are not shared across channels.
    • Range of values:
      • 0 - do not share mean values across channels
      • 1 - share mean values across channels
    • Type: int
    • Default value: 0
    • Required: no
  • Parameter name: normalize_variance
    • Description: normalize_variance is a flag that specifies whether to perform variance normalization.
    • Range of values:
      • 0 - do not normalize variance
      • 1 - normalize variance
    • Type: int
    • Default value: 0
    • Required: no
  • Parameter name: eps
    • Description: eps is the number to be added to the variance to avoid division by zero when normalizing the value. For example, epsilon equal to 0.001 means that 0.001 is added to the variance.
    • Range of values: a positive floating-point number
    • Type: float
    • Default value: None
    • Required: yes

Inputs

  • 1: 4D or 5D input blob. Required.

Mathematical Formulation

MVN subtracts mean value from the input blob:

\[ o_{i} = i_{i} - \frac{\sum{i_{k}}}{C * H * W} \]

If normalize_variance is set to 1, the output blob is divided by variance:

\[ o_{i}=\frac{o_{i}}{\sum \sqrt {o_{k}^2}+\epsilon} \]

Example

<layer ... type="MVN">
<data across_channels="1" eps="9.999999717180685e-10" normalize_variance="1"/>
<input>
...
</input>
<output>
...
</output>
</layer>

NonMaxSuppression Layer

Back to top

Name: NonMaxSuppression

Category: Layer

Short description: NonMaxSuppression performs non-maximum suppression of the input boxes and return indices of the selected boxes.

Detailed description: Reference

Parameters: NonMaxSuppression layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: center_point_box
    • Description: center_point_box is flag that specifies the format of the box data.
    • Range of values:
      • false (0) - the box data is supplied as [y1, x1, y2, x2] where (y1, x1) and (y2, x2) are the coordinates of any diagonal pair of box corners.
      • true (1) - the box data is supplied as [x_center, y_center, width, height].
    • Type: bool
    • Default value: false
    • Required: no

Inputs

  • 1: 3D floating point blob with the boxes data of shape [batch_size, num_boxes, 4]. Required.
  • 2: 3D floating point blob with the boxes scores of shape [batch_size, num_classes, num_boxes]. Required.
  • 3: 1D integer blob with of shape [1] representing maximum number of boxes to be selected per class. Optional. If not specified then all boxes will be selected.
  • 4: 1D floating point blob with of shape [1] representing intersection over union threshold. Optional. If not specified then it is equal to 1.0.
  • 5: 1D floating point blob with of shape [1] representing box score threshold. Optional. If not specified then it is equal to 0.0.

Mathematical Formulation

\[o_{i} = \left( 1 + \left( \frac{\alpha}{n} \right)\sum_{i}x_{i}^{2} \right)^{\beta}\]

Where \(n\) is the size of each local region.

Example

<layer ... type="Norm" ... >
<data alpha="9.9999997e-05" beta="0.75" local-size="5" region="across"/>
<input> ... </input>
<output> ... </output>
</layer>

Norm Layer

Back to top

Name: Norm

Category: Normalization

Short description: Reference

Detailed description: Reference

Parameters: Norm layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: alpha
    • Description: alpha is a scaling parameter for the normalizing sum. For example, alpha equal to 0.0001 means that the normalizing sum is multiplied by 0.0001.
    • Range of values: a positive floating-point number
    • Type: float
    • Default value: None
    • Required: yes
  • Parameter name: beta
    • Description: beta is an exponent for the normalizing sum. For example, beta equal to 0.75 means that the normalizing sum is raised to the power of 0.75.
    • Range of values: a positive floating-point number
    • Type: float
    • Default value: None
    • Required: yes
  • Parameter name: region
    • Description: region is the strategy of local regions extension. For example, region equal to across means that the normalizing sum is performed over adjacent channels.
    • Range of values:
      • across - normalize sum over adjacent channels
      • same - normalize sum over nearby spatial locations
    • Type: string
    • Default value: across
    • Required: yes
  • Parameter name: local-size
    • Description: local-size represents the side length of the region to be used for the normalization sum or number of channels depending on the strategy specified in the region parameter. For example, local-size equal to 5 for the across strategy means application of sum across 5 adjacent channels.
    • Range of values: a positive integer
    • Type: int
    • Default value: None
    • Required: yes

Inputs

  • 1: 4D input blob. Required.

Mathematical Formulation

\[o_{i} = \left( 1 + \left( \frac{\alpha}{n} \right)\sum_{i}x_{i}^{2} \right)^{\beta}\]

Where \(n\) is the size of each local region.

Example

<layer ... type="Norm" ... >
<data alpha="9.9999997e-05" beta="0.75" local-size="5" region="across"/>
<input> ... </input>
<output> ... </output>
</layer>

Normalize Layer

Back to top

Name: Normalize

Category: Normalization

Short description: Normalize layer performs l-p normalization of 1 of input blob.

Parameters: Normalize layer parameters should be specified as the data node, which is a child of the layer node.

  • Parameter name: across_spatial
    • Description: across_spatial is a flag that specifies if normalization is performed over CHW or HW. For example, across_spatial equal to 0 means that normalization is not shared across channels.
    • Range of values:
      • 0 - do not share normalization across channels
      • 1 - not supported
    • Type: int
    • Default value: 0
    • Required: no
  • Parameter name: channel_shared
    • Description: channel_shared is a flag that specifies if scale parameters are shared across channels. For example, channel_shared equal to 0 means that scale parameters are not shared across channels.
    • Range of values:
      • 0 - do not share scale parameters across channels
      • 1 - not supported
    • Type: int
    • Default value: 0
    • Required: no
  • Parameter name: eps
    • Description: eps is the number to be added to the variance to avoid division by zero when normalizing the value. For example, eps equal to 0.001 means that 0.001 is used if all the values in normalization are equal to zero.
    • Range of values: a positive floating-point number
    • Type: float
    • Default value: None
    • Required: yes

Inputs

  • 1: 2D, 3D or 4D input blob. Required.

Mathematical Formulation

\[ o_{i} = \sum_{i}^{H*W}\frac{\left ( n*C*H*W \right )` scale}{\sqrt{\sum_{i=0}^{C*H*W}\left ( n*C*H*W \right )^{2}}} \]

Example

<layer ... type="Normalize" ... >
<data across_spatial="0" channel_shared="0" eps="0.000000"/>
<input> ... </input>
<output> ... </output>
</layer>

OneHot Layer

Back to top

Name: OneHot

Category: Layer

Short description: OneHot layer fills the locations represented by indices specified in input with the value of on_value and fills all other locations with the value of off_value. If an index is out of range, the corresponding element is also filled with the off_value.

Detailed description: Reference

Parameters: OneHot layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: axis
    • Description: axis is a new axis position in the output shape to fill with one-hot values.
    • Range of values: an integer. Negative value means counting dimension from the end.
    • Type: int
    • Default value: -1
    • Required: no
  • Parameter name: depth
    • Description: depth is depth of a new one-hot dimension.
    • Range of values: a non-negative integer
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: on_value
    • Description: on_value is the value that the locations represented by indices in input take.
    • Range of values: a floating-point number.
    • Type: float
    • Default value: 1.0
    • Required: no
  • Parameter name: off_value
    • Description: off_value is the value that the locations not represented by indices in input take.
    • Range of values: a floating-point number.
    • Type: float
    • Default value: 0.0
    • Required: no

Inputs:

  • 1: Multidimensional input tensor with indices of type T (can be 0D). Required.

Outputs:

  • 1 Multidimensional output tensor. If the input indices have rank N, the output will have rank N+1. A new axis of the size depth is created at the dimension axis.

Examples

<layer ... type="OneHot" ...>
<data axis="-1" depth="3" on_value="1.0" off_value="0.0"/>
<input>
<port id="0"> <!-- indices value [0, 1, 2] -->
<dim>3</dim>
</port>
</input>
<output>
<port id="2"> <!-- output value # [[1., 0., 0.], [0., 1., 0.], [0., 0., 1.]] -->
<dim>3</dim>
<dim>3</dim>
</port>
</output>
</layer>

Pad Layer

Back to top

Name: Pad

Category: Layer

Short description: Pad layer extends an input blob on edges. New element values are generated based on the Pad layer parameters described below.

Parameters: Pad layer parameters are specified in the data section, which is a child of the layer node. The parameters specify a number of elements to add along each axis and a rule by which new element values are generated: for example, whether they are filled with a given constant or generated based on the input blob content.

  • Parameter name: pads_begin
    • Description: pads_begin specifies the number of padding elements at the beginning of each axis.
    • Range of values: a list of non-negative integers. The length of the list must be equal to the number of dimensions in the input blob.
    • Type: int[]
    • Default value: None
    • Required: yes
  • Parameter name: pads_end
    • Description: pads_end specifies the number of padding elements at the end of each axis.
    • Range of values: a list of non-negative integers. The length of the list must be equal to the number of dimensions in the input blob.
    • Type: int[]
    • Default value: None
    • Required: yes
  • Parameter name: pad_mode
    • Description: pad_mode specifies the method used to generate new element values.
    • Range of values: Name of the method in string format:
      • constant - padded values are equal to the value of the pad_value layer parameter.
      • edge - padded values are copied from the respective edge of the input blob.
      • reflect - padded values are a reflection of the input blob; values on the edges are not duplicated. pads_begin[D] and pads_end[D] must be not greater than input.shape[D] – 1 for any valid D.
      • symmetric - padded values are symmetrically added from the input blob. This method is similar to the reflect, but values on edges are duplicated. Refer to the examples below for more details. pads_begin[D] and pads_end[D] must be not greater than input.shape[D] for any valid D.
    • Type: string
    • Default value: None
    • Required: yes
  • Parameter name: pad_value
    • Description: Use with the pad_mode = "constant" only. All new elements are filled with this value.
    • Range of values: a floating-point number
    • Type: float
    • Default value: 0.0
    • Required: no

Inputs

  • 1: Multidimensional input blob. Required.

Outputs

  • 1: Multidimensional input blob with dimensions pads_begin[D] + input.shape[D] + pads_end[D] for each D from 0 to len(input.shape) - 1.

pad_mode Examples

The following examples illustrate how output blob is generated for the Pad layer for a given input blob:

INPUT =
[[ 1 2 3 4 ]
[ 5 6 7 8 ]
[ 9 10 11 12 ]]

with the following parameters:

pads_begin = [0, 1]
pads_end = [2, 3]

depending on the pad_mode.

  • pad_mode = "constant":
    OUTPUT =
    [[ 0 1 2 3 4 0 0 0 ]
    [ 0 5 6 7 8 0 0 0 ]
    [ 0 9 10 11 12 0 0 0 ]
    [ 0 0 0 0 0 0 0 0 ]
    [ 0 0 0 0 0 0 0 0 ]]
  • pad_mode = "edge":
    OUTPUT =
    [[ 1 1 2 3 4 4 4 4 ]
    [ 5 5 6 7 8 8 8 8 ]
    [ 9 9 10 11 12 12 12 12 ]
    [ 9 9 10 11 12 12 12 12 ]
    [ 9 9 10 11 12 12 12 12 ]]
  • pad_mode = "reflect":
    OUTPUT =
    [[ 2 1 2 3 4 3 2 1 ]
    [ 6 5 6 7 8 7 6 5 ]
    [ 10 9 10 11 12 11 10 9 ]
    [ 6 5 6 7 8 7 6 5 ]
    [ 2 1 2 3 4 3 2 1 ]]
  • pad_mode = "symmetric":
    OUTPUT =
    [[ 1 1 2 3 4 4 3 2 ]
    [ 5 5 6 7 8 8 7 6 ]
    [ 9 9 10 11 12 12 11 10 ]
    [ 9 9 10 11 12 12 11 10 ]
    [ 5 5 6 7 8 8 7 6 ]]

Example

<layer ... type="Pad" ...>
<data pads_begin="0,5,2,1" pads_end="1,0,3,7" pad_mode="constant" pad_value="666.0"/>
<input>
<port id="0">
<dim>1</dim>
<dim>3</dim>
<dim>32</dim>
<dim>40</dim>
</port>
</input>
<output>
<port id="2">
<dim>2</dim> <!-- 2 = 0 + 1 + 1 = pads_begin[0] + input.shape[0] + pads_end[0] -->
<dim>8</dim> <!-- 8 = 5 + 3 + 0 = pads_begin[1] + input.shape[1] + pads_end[1] -->
<dim>37</dim> <!-- 37 = 2 + 32 + 3 = pads_begin[2] + input.shape[2] + pads_end[2] -->
<dim>48</dim> <!-- 48 = 1 + 40 + 7 = pads_begin[3] + input.shape[3] + pads_end[3] -->
</port>
</output>
</layer>

Permute Layer

Back to top

Name: Permute

Category: Layer

Short description: Permute layer reorders input blob dimensions.

Detailed description: Reference

Parameters: Permute layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: order
    • Description: order is a list of dimensions indexes for output blob. For example, order equal to "0,2,3,1" means that the output blob has the following dimensions: the first dimension from the input blob, the third dimension from the input blob, the fourth dimension from the input blob, the second dimension from the input blob.
    • Range of values: a list of positive integers separated by comma
    • Type: int[]
    • Default value: None
    • Required: yes

Inputs:

  • 1: Multidimensional input blob. Required.

Mathematical Formulation

Permute layer reorders input blob dimensions. Source indexes and destination indexes are bound by the formula:

\[ src\_ind_{offset} = n * ordered[1] * ordered[2] * ordered[3] + (h * ordered[3] + w) \]

\[ n \in ( 0, order[0] ) \]

\[ h \in ( 0, order[2] ) \]

\[ w \in ( 0, order[3] ) \]

Example

<layer ... type="Permute" ... >
<data order="0,2,3,1"/>
<input> ... </input>
<output> ... </output>
</layer>

Pooling Layer

Back to top

Name: Pooling

Category: Pool

Short description: Reference

Detailed description: Reference

Parameters: Pooling layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: strides
    • Description: strides is a distance (in pixels) to slide the window on the feature map over the (z, y, x) axes for 3D poolings and (y, x) axes for 2D poolings. For example, strides equal to "4,2,1" means sliding the window four pixels at a time over depth dimension, two pixels over height dimension, and one pixel over width dimension.
    • Range of values: a list of non-negative integers
    • Type: int[]
    • Default value: a list of 1 with length equal to the number of convolution kernel dimensions
    • Required: no
  • Parameter name: pads_begin
    • Description: pads_begin is the number of pixels to add to the beginning along each axis. For example, pads_begin equal to "1,2" means adding one pixel to the top of the input and two pixels to the left of the input.
    • Range of values: a list of non-negative integers
    • Type: int[]
    • Default value: a list of 0 with length equal to the number of convolution kernel dimensions
    • Required: no
  • Parameter name: pads_end
    • Description: pads_end is the number of pixels to add to the ending along each axis. For example, pads_end equal "1,2" means adding one pixel to the bottom of the input and two pixels to the right of the input.
    • Range of values: a list of non-negative integers
    • Type: int[]
    • Default value: a list of 1 with length equal to the number of convolution kernel dimensions
    • Required: no
  • Parameter name: kernel
    • Description: kernel is a size of each filter. For example, kernel equal to "2,3" means that each filter has height equal to 2 and width equal to 3.
    • Range of values: a list of positive integers
    • Type: int[]
    • Default value: None
    • Required: yes
  • Parameter name: pool-method
    • Description: pool-method is a type of pooling strategy for values.
    • Range of values:
      • max - choose the biggest value in a feature map for each window position
      • avg - take the average value in a feature map for each windows position
    • Type: string
    • Default value: None
    • Required: yes
  • Parameter name: exclude-pad
    • Description: exclude-pad is a flag that specifies whether to ignore zeros in a padding area. For example, exclude-pad equal to true means that zero values in the padding are not used.
    • Range of values: true or false
    • Type: string
    • Default value: None
    • Required: yes
  • Parameter name: rounding_type
    • Description: rounding_type is a type of rounding to apply.
    • Range of values:
      • ceil
      • floor
    • Type: string
    • Default value: floor
    • Required: no
  • Parameter name: auto_pad
    • Description: auto_pad specifies how to calculate padding.
    • Range of values:
      • Not specified: use explicit padding values
      • same_upper/same_lower: the input is padded to match the output size. In case of odd padding value, an extra padding is added at the end (at the beginning).
      • valid: do not use padding
    • Type: string
    • Default value: None
    • Required: no

Inputs:

  • 1: 4D or 5D input blob. Required.

Mathematical Formulation

  • For pool-method="max":

    \[ output_{j} = MAX\{ x_{0}, ... x_{i}\} \]

  • For pool-method="avg":

    \[ output_{j} = \frac{\sum_{i = 0}^{n}x_{i}}{n} \]

Example

<layer ... type="Pooling" ... >
<data auto_pad="same_upper" exclude-pad="true" kernel="3,3" pads_begin="0,0" pads_end="1,1" pool-method="max" strides="2,2"/>
<input> ... </input>
<output> ... </output>
</layer>

Power Layer

Back to top

Name: Power

Category: Layer

Short description: Power layer computes the output as (shift + scale * x) ^ power for each input element x.

Parameters: Power layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: power
    • Description: power is a parameter in the formula described above.
    • Range of values: a floating-point number
    • Type: float
    • Default value: None
    • Required: yes
  • Parameter name: scale
    • Description: scale is a parameter in the formula described above.
    • Range of values: a floating-point number
    • Type: float
    • Default value: None
    • Required: yes
  • Parameter name: shift
    • Description: shift is a parameter in the formula described above.
    • Range of values: a floating-point number
    • Type: int
    • Default value: None
    • Required: yes

Inputs:

  • 1: Multidimensional input blob. Required.

Mathematical Formulation

\[ p = (shift + scale * x)^{power} \]

Example

<layer ... type="Power" ... >
<data power="2" scale="0.1" shift="5"/>
<input> ... </input>
<output> ... </output>
</layer>

PReLU Layer

Back to top

Name: PReLU

Category: Activation

Short description: PReLU is the Parametric Rectifier Linear Unit. The difference from ReLU is that negative slopes can vary across channels.

Parameters: PReLU layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: channel_shared
    • Description: channel_shared specifies whether a negative slope is shared across channels or not. If the channel_shared is equal to 0, the slope shape is equal to the number of channels, if the channel_shared is equal to 1, the slope is scalar.
    • Range of values: 0 or 1
    • Type: int
    • Default value: 0
    • Required: no

Inputs:

  • 1: 4D or 5D input blob. Required.

Mathematical Formulation

PReLU accepts one input with four dimensions. The produced blob has the same dimensions as input. PReLU does the following with the input blob:

\[ o_{i} = max(0, x_{i}) + w_{i} * min(0,x_{i}) \]

where \(w_{i}\) is from weights blob.

Example

<layer ... type="PReLU" ... >
<data channel_shared="1"/>
<input> ... </input>
<output> ... </output>
</layer>

PriorBox Layer

Back to top

Name: PriorBox

Category: Layer

Short description: PriorBox layer generates prior boxes of specified sizes and aspect ratios across all dimensions.

Parameters: PriorBox layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: min_size
    • Description: min_size is the minimum box size (in pixels). For example, min_size equal to [15.0] means that the minimum box size is 15.0.
    • Range of values: a list of positive floating-point numbers
    • Type: float[]
    • Default value: []
    • Required: yes
  • Parameter name: max_size
    • Description: max_size is the maximum box size (in pixels). For example, max_size equal to [15.0] means that the maximum box size is 15.0.
    • Range of values: a list of positive floating-point numbers
    • Type: float[]
    • Default value: []
    • Required: yes
  • Parameter name: aspect_ratio
    • Description: aspect_ratio is a variance of aspect ratios. Duplicate values are ignored. For example, aspect_ratio equal to "[2.0,3.0]" means that for the first box, aspect ratio is 2.0, for the second box, it is 3.0.
    • Range of values: a list of positive floating-point numbers
    • Type: float[]
    • Default value: []
    • Required: no
  • Parameter name: flip
    • Description: flip is a flag that specifies whether each aspect_ratio is duplicated and flipped. For example, flip equal to 1 and aspect_ratio equal to "4.0,2.0" mean that aspect_ratio is equal to "4.0,2.0,0.25,0.5".
    • Range of values:
      • 0 - flip each aspect_ratio
      • 1 - do not flip each aspect_ratio
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: clip
    • Description: clip is a flag that specifies if each value in the output blob is clipped to [0,1] interval.
    • Range of values:
      • 0 - do not perform clipping
      • 1 - clip each value in the output blob to [0,1] interval
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: step
    • Description: step is the distance between box centers. For example, step equal to 85.0 means that the distance between neighborhood prior boxes centers is 85.0.
    • Range of values: a non-negative floating-point number
    • Type: float
    • Default value: 0.0
    • Required: yes
  • Parameter name: offset
    • Description: offset is a shift of box respectively to top left corner. For example, offset equal to 85.0 means that the shift of neighborhood prior boxes centers is 85.0.
    • Range of values: a non-negative floating-point number
    • Type: float
    • Default value: None
    • Required: yes
  • Parameter name: variance
    • Description: variance is the variance of adjusting bounding boxes. The parameter can contain 0, 1 or 4 elements.
    • Range of values: a list of positive floating-point numbers
    • Type: float[]
    • Default value: []
    • Required: yes
  • Parameter name: scale_all_sizes
    • Description: scale_all_sizes is a flag that specifies the type of inference. For example, scale_all_sizes equal to 0 means that the PriorBox layer is inferred in MXNet-like manner, which means that the max_size parameter is ignored.
    • Range of values:
      • 0 - do not use max_size
      • 1 - use max_size
    • Type: int
    • Default value: 1
    • Required: no
  • Parameter name: fixed_ratio
    • Description: fixed_ratio is an aspect ratio of a box. For example, fixed_ratio equal to 2.000000 means that the aspect ratio for the first box aspect ratio is 2.
    • Range of values: a list of positive floating-point numbers
    • Type: float[]
    • Default value: None
    • Required: no
  • Parameter name: fixed_size
    • Description: fixed_size is an initial box size (in pixels). For example, fixed_size equal to 15 means that the initial box size is 15.
    • Range of values: a list of positive floating-point numbers
    • Type: float[]
    • Default value: None
    • Required: no
  • Parameter name: density
    • Description: density is the square root of the number of boxes of each type. For example, density equal to 2 means that the first box generates four boxes of the same size and with the same shifted centers.
    • Range of values: a list of positive floating-point numbers
    • Type: float[]
    • Default value: None
    • Required: no

Inputs:

  • 1: 4D input blob. Used to get height and width only. Required.
  • 2: 4D input blob. Used to get image height and image width only. Required.

Mathematical Formulation:

PriorBox computes coordinates of prior boxes as follows:

  1. Calculates center_x and center_y of prior box:

    \[ W \equiv Width \quad Of \quad Image \]

    \[ H \equiv Height \quad Of \quad Image \]

    • If step equals 0:

      \[ center_x=(w+0.5) \]

      \[ center_y=(h+0.5) \]

    • else:

      \[ center_x=(w+offset)`step \]

      \[ center_y=(h+offset)`step \]

      \[ w \subset \left( 0, W \right ) \]

      \[ h \subset \left( 0, H \right ) \]

  2. For each \( s \subset \left( 0, min_sizes \right ) \), calculates coordinates of prior boxes:

    \[ xmin = \frac{\frac{center_x - s}{2}}{W} \]

    \[ ymin = \frac{\frac{center_y - s}{2}}{H} \]

    \[ xmax = \frac{\frac{center_x + s}{2}}{W} \]

    \[ ymax = \frac{\frac{center_y + s}{2}}{H} \]

Example

<layer ... type="PriorBox" ... >
<data step="64.000000" min_size="162.000000" max_size="213.000000" offset="0.500000" flip="1" clip="0" aspect_ratio="2.000000,3.000000" variance="0.100000,0.100000,0.200000,0.200000" />
<input> ... </input>
<output> ... </output>
</layer>

PriorBoxClustered Layer

Back to top

Name: PriorBoxClustered

Category: Layer

Short description: PriorBoxClustered layer generates prior boxes of specified sizes normalized to the input image size.

Parameters: PriorBoxClustered layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: width
    • Description: width specifies desired boxes widths in pixels.
    • Range of values: a list of positive floating-point numbers
    • Type: float[]
    • Default value: []
    • Required: yes
  • Parameter name: height
    • Description: height specifies desired boxes heights in pixels.
    • Range of values: positive floating-point numbers
    • Type: float[]
    • Default value: []
    • Required: yes
  • Parameter name: clip
    • Description: clip is a flag that specifies if each value in the output blob is clipped within [0,1].
    • Range of values:
      • 0 - do not perform clipping
      • 1 - clip each value in the output blob to [0,1]
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: step (step_w, step_h)
    • Description: step (step_w, step_h) is the distance between box centers. For example, step equal to 85.0 means that the distance between neighborhood prior boxes centers is 85.0. If both step_h and step_w are 0.0, they are updated with value of step. If after that they are still 0.0, they are calculated as input image heights/width divided by the first input heights/width.
    • Range of values: a positive floating-point number
    • Type: float
    • Default value: 0.0
    • Required: yes
  • Parameter name: offset
    • Description: offset is a shift of box respectively to top left corner. For example, offset equal to 85.0 means that the shift of neighborhood prior boxes centers is 85.0.
    • Range of values: a positive floating-point number
    • Type: float
    • Default value: None
    • Required: yes
  • Parameter name: variance
    • Description: variance is the variance of adjusting bounding boxes.
    • Range of values: a list of positive floating-point numbers
    • Type: float[]
    • Default value: []
    • Required: yes
  • Parameter name: img_h
    • Description: img_h is the height of input image. It is calculated as the second input height unless provided explicitly.
    • Range of values: a positive floating-point number
    • Type: float
    • Default value: 1.0
    • Required: yes
  • Parameter name: img_w
    • Description: img_w is the width of input image. It is calculated as second input width unless provided explicitly.
    • Range of values: a positive floating-point number
    • Type: float
    • Default value: 1.0
    • Required: yes

Inputs:

  • 1: 4D input blob. Used to get height and width only. Required.
  • 2: 4D input blob. Used to get image height and image width only. Required.

Mathematical Formulation

PriorBoxClustered computes coordinates of prior boxes as follows:

  1. Calculates the center_x and center_y of prior box:

    \[ W \equiv Width \quad Of \quad Image \]

    \[ H \equiv Height \quad Of \quad Image \]

    \[ center_x=(w+offset)`step \]

    \[ center_y=(h+offset)`step \]

    \[ w \subset \left( 0, W \right ) \]

    \[ h \subset \left( 0, H \right ) \]

  2. For each \(s \subset \left( 0, W \right )\), calculates the prior boxes coordinates:

    \[ xmin = \frac{center_x - \frac{width_s}{2}}{W} \]

    \[ ymin = \frac{center_y - \frac{height_s}{2}}{H} \]

    \[ xmax = \frac{center_x - \frac{width_s}{2}}{W} \]

    \[ ymax = \frac{center_y - \frac{height_s}{2}}{H} \]

    If clip is defined, the coordinates of prior boxes are recalculated with the formula: \(coordinate = \min(\max(coordinate,0), 1)\)

Example

<layer ... type="PriorBoxClustered">
<data clip="0" flip="0" height="44.0,10.0,30.0,19.0,94.0,32.0,61.0,53.0,17.0" offset="0.5" step="16.0" variance="0.1,0.1,0.2,0.2"
width="86.0,13.0,57.0,39.0,68.0,34.0,142.0,50.0,23.0"/>
<input>
...
</input>
<output>
...
</output>
</layer>

Proposal Layer

Back to top

Name: Proposal

Category: Layer

Short description: Proposal layer filters bounding boxes and outputs only those with the highest prediction confidence.

Parameters: Proposal layer parameters are specified in the data node, which is a child of the layer node. The layer has three inputs: a blob with probabilities whether particular bounding box corresponds to background and foreground, a blob with logits for each of the bounding boxes, a blob with input image size in the [image_height, image_width, scale_height_and_width] or [image_height, image_width, scale_height, scale_width] format.

  • Parameter name: base_size
    • Description: base_size is the size of the anchor to which scale and ratio parameters are applied.
    • Range of values: a positive integer number
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: pre_nms_topn
    • Description: pre_nms_topn is the number of bounding boxes before the NMS operation. For example, pre_nms_topn equal to 15 means that the minimum box size is 15.
    • Range of values: a positive integer number
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: post_nms_topn
    • Description: post_nms_topn is the number of bounding boxes after the NMS operation. For example, post_nms_topn equal to 15 means that the maximum box size is 15.
    • Range of values: a positive integer number
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: nms_thresh
    • Description: nms_thresh is the minimum value of the proposal to be taken into consideration. For example, nms_thresh equal to 0.5 means that all boxes with prediction probability less than 0.5 are filtered out.
    • Range of values: a positive floating-point number
    • Type: float
    • Default value: None
    • Required: yes
  • Parameter name: feat_stride
    • Description: feat_stride is the step size to slide over boxes (in pixels). For example, feat_stride equal to 16 means that all boxes are analyzed with the slide 16.
    • Range of values: a positive integer
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: min_size
    • Description: min_size is the minimum size of box to be taken into consideration. For example, min_size equal 35 means that all boxes with box size less than 35 are filtered out.
    • Range of values: a positive integer number
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: ratio
    • Description: ratio is the ratios for anchor generation.
    • Range of values: a list of floating-point numbers
    • Type: float[]
    • Default value: []
    • Required: yes
  • Parameter name: scale
    • Description: scale is the scales for anchor generation.
    • Range of values: a list of floating-point numbers
    • Type: float[]
    • Default value: []
    • Required: yes
  • Parameter name: clip_before_nms
    • Description: clip_before_nms flag that specifies whether to perform clip bounding boxes before non-maximum suppression or not.
    • Range of values: 0 or 1
    • Type: int
    • Default value: 1
    • Required: no
  • Parameter name: clip_after_nms
    • Description: clip_after_nms is a flag that specifies whether to perform clip bounding boxes after non-maximum suppression or not.
    • Range of values: 0 or 1
    • Type: int
    • Default value: 0
    • Required: no
  • Parameter name: normalize
    • Description: normalize is a flag that specifies whether to perform normalization of output boxes to [0,1] interval or not.
    • Range of values: 0 or 1
    • Type: int
    • Default value: 0
    • Required: no
  • Parameter name: box_size_scale
    • Description: box_size_scale specifies the scale factor applied to logits of box sizes before decoding.
    • Range of values: a positive floating-point number
    • Type: float
    • Default value: 1.0
    • Required: no
  • Parameter name: box_coordinate_scale
    • Description: box_coordinate_scale specifies the scale factor applied to logits of box coordinates before decoding.
    • Range of values: a positive floating-point number
    • Type: float
    • Default value: 1.0
    • Required: no
  • Parameter name: framework
    • Description: framework specifies how the box coordinates are calculated.
    • Range of values:
      • "" (empty string) - calculate box coordinates like in Caffe*
      • tensorflow - calculate box coordinates like in the TensorFlow* Object Detection API models
    • Type: string
    • Default value: "" (empty string)
    • Required: no
  • Parameter name: for_deformable
    • Description: for_deformable specifies how the box coordinates are calculated.
    • Range of values: 0 or 1
    • Type: int
    • Default value: 0
    • Required: no

Mathematical Formulation

Proposal layer accepts three inputs with four dimensions. The produced blob has two dimensions: the first one equals batch_size * post_nms_topn. Proposal layer does the following with the input blob:

  1. Generates initial anchor boxes. Left top corner of all boxes is at (0, 0). Width and height of boxes are calculated from base_size with scale and ratio parameters.
  2. For each point in the first input blob:
    • pins anchor boxes to the image according to the second input blob that contains four deltas for each box: for x and y of center, for width and for height
    • finds out score in the first input blob
  3. Filters out boxes with size less than min_size
  4. Sorts all proposals (box, score) by score from highest to lowest
  5. Takes top pre_nms_topn proposals
  6. Calculates intersections for boxes and filter out all boxes with \(intersection/union > nms\_thresh\)
  7. Takes top post_nms_topn proposals
  8. Returns top proposals

Inputs:

  • 1: 4D input blob with class prediction scores. Required.
  • 2: 4D input blob with box logits. Required.
  • 3: 1D input blob 3 or 4 elements: [image height, image width, scale for image height/width OR scale for image height and scale for image width]. Required.

Example

<layer ... type="Proposal" ... >
<data base_size="16" feat_stride="16" min_size="16" nms_thresh="0.6" post_nms_topn="200" pre_nms_topn="6000"
ratio="2.67" scale="4.0,6.0,9.0,16.0,24.0,32.0"/>
<input> ... </input>
<output> ... </output>
</layer>

PSROIPooling Layer

Back to top

Name: PSROIPooling

Category: Pool

Short description: PSROIPooling layer compute position-sensitive pooling on regions of interest specified by input.

Detailed description: Reference

Parameters: PSRoiPooling layer parameters are specified in the data node, which is a child of the layer node. PSROIPooling layer takes two input blobs: with feature maps and with regions of interests (box coordinates). The latter is specified as five element tuples: [batch_id, x_1, y_1, x_2, y_2]. ROIs coordinates are specified in absolute values for the average mode and in normalized values (to [0,1] interval) for bilinear interpolation.

  • Parameter name: output_dim
    • Description: output_dim is a pooled output channel number.
    • Range of values: a positive integer
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: group_size
    • Description: group_size is the number of groups to encode position-sensitive score maps. Use for average mode only.
    • Range of values: a positive integer
    • Type: int
    • Default value: 1
    • Required: no
  • Parameter name: spatial_scale
    • Description: spatial_scale is a multiplicative spatial scale factor to translate ROI coordinates from their input scale to the scale used when pooling.
    • Range of values: a positive floating-point number
    • Type: float
    • Default value: None
    • Required: yes
  • Parameter name: mode
    • Description: mode specifies mode for pooling.
    • Range of values:
      • average - perform average pooling
      • bilinear - perform pooling with bilinear interpolation
    • Type: string
    • Default value: average
    • Required: yes
  • Parameter name: spatial_bins_x
    • Description: spatial_bins_x specifies numbers of bins to divide the input feature maps over width. Used for "bilinear" mode only.
    • Range of values: a positive integer
    • Type: int
    • Default value: 1
    • Required: no
  • Parameter name: spatial_bins_y
    • Description: spatial_bins_y specifies numbers of bins to divide the input feature maps over height. Used for bilinear mode only.
    • Range of values: a positive integer
    • Type: int
    • Default value: 1
    • Required: no

Inputs:

  • 1: 4D input blob with feature maps. Required.
  • 2: 2D input blob describing box consisting of five element tuples: [batch_id, x_1, y_1, x_2, y_2]. Required.

Example

<layer ... type="PSROIPooling" ... >
<data group_size="6" mode="bilinear" output_dim="360" spatial_bins_x="3" spatial_bins_y="3" spatial_scale="1"/>
<input>
<port id="0">
<dim>1</dim>
<dim>3240</dim>
<dim>38</dim>
<dim>38</dim>
</port>
<port id="1">
<dim>100</dim>
<dim>5</dim>
</port>
</input>
<output>
<port id="2">
<dim>100</dim>
<dim>360</dim>
<dim>6</dim>
<dim>6</dim>
</port>
</output>
</layer>

FakeQuantize Layer

Back to top

Name: FakeQuantize

Category: Layer

Short description: FakeQuantize layer is element-wise linear quantization of floating-point input values into a discrete set of floating-point values.

Detailed description: Input and output ranges as well as the number of levels of quantization are specified by dedicated inputs and attributes. There can be different limits for each element or groups of elements (channels) of the input blobs. Otherwise, one limit applies to all elements. It depends on shape of inputs that specify limits and regular broadcasting rules applied for input blobs. The output of the operator is a floating-point number of the same type as the input blob. In general, there are four values that specify quantization for each element: input_low, input_high, output_low, output_high. input_low and input_high parameters specify the input range of quantization. All input values that are outside this range are clipped to the range before actual quantization. output_low and output_high specify minimum and maximum quantized values at the output.

Parameters: Quantize layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: levels
    • Description: levels is the number of quantization levels.
    • Range of values: an integer greater than or equal to 2
    • Type: int
    • Default value: None
    • Required: yes

Inputs:

  • 1: X - multidimensional input blob to quantize. Required.
  • 2: input_low - minimum limit for input value. The shape must be broadcastable to the shape of X. Required.
  • 3: input_high - maximum limit for input value. Can be the same as input_low for binarization. The shape must be broadcastable to the shape of X. Required.
  • 4: output_low - minimum quantized value. The shape must be broadcastable to the shape of X. Required.
  • 5: output_high - maximum quantized value. The shape must be broadcastable to the of X. Required.

Mathematical Formulation

Each element of the output is defined as the result of the following expression:

if x <= input_low:
output = output_low
elif x > input_high:
output = output_high
else:
# input_low < x <= input_high
output = round((x - input_low) / (input_high - input_low) * (levels-1)) / (levels-1) * (output_high - output_low) + output_low

Example

<layer type="FakeQuantize">
<data levels="2"/>
<input>
<port id="0">
<dim>1</dim>
<dim>64</dim>
<dim>56</dim>
<dim>56</dim>
</port>
<port id="1">
<dim>1</dim>
<dim>64</dim>
<dim>1</dim>
<dim>1</dim>
</port>
<port id="2">
<dim>1</dim>
<dim>64</dim>
<dim>1</dim>
<dim>1</dim>
</port>
<port id="3">
<dim>1</dim>
<dim>1</dim>
<dim>1</dim>
<dim>1</dim>
</port>
<port id="4">
<dim>1</dim>
<dim>1</dim>
<dim>1</dim>
<dim>1</dim>
</port>
</input>
<output>
<port id="5">
<dim>1</dim>
<dim>64</dim>
<dim>56</dim>
<dim>56</dim>
</port>
</output>
</layer>

Range Layer

Back to top

Name: Range

Category: Layer

Short description: Range sequence of numbers according input values.

Detailed description: Range layers generates a sequence of numbers starting from the value in the first input up to but not including the value in the second input with a step equal to the value in the third input.

Parameters: Range layer does not have parameters.

Inputs:

  • 1: 0D blob (constant) with the start value of the range. Required.
  • 2: 0D blob (constant) with the limit value of the range. Required.
  • 3: 0D blob (constant) with the step value. Required.

Example

<layer ... type="Range">
<input>
<port id="0"/> <!-- value 5-->
<port id="1"/> <!-- value 15-->
<port id="2"/> <!-- value 1-->
</input>
<output>
<port id="3">
<dim>10</dim> <!-- output value is: [5,6,7,8,9,10,11,12,13,14]-->
</port>
</output>
</layer>

RegionYolo Layer

Back to top

Name: RegionYolo

Category: Layer

Short description: RegionYolo computes the coordinates of regions with probability for each class.

Detailed description: Reference

Parameters: RegionYolo layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: coords
    • Description: coords is the number of coordinates for each region.
    • Range of values: an integer
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: classes
    • Description: classes is the number of classes for each region.
    • Range of values: an integer
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: num
    • Description: num is the number of regions.
    • Range of values: an integer
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: do_softmax
    • Description: do_softmax is a flag that specifies the inference method and affects how the number of regions is determined.
    • Range of values:
      • 0 - do not perform softmax
      • 1 - perform softmax
    • Type: int
    • Default value: 1
    • Required: no
  • Parameter name: mask
    • Description: mask specifies the number of regions. Use this parameter instead of num when do_softmax is equal to 0.
    • Range of values: a list of integers
    • Type: int[]
    • Default value: []
    • Required: no

Inputs:

  • 1: 4D input blob. Required.

Example

<layer ... type="RegionYolo" ... >
<data axis="1" classes="80" coords="4" do_softmax="0" end_axis="3" mask="0,1,2" num="9"/>
<input> ... </input>
<output> ... </output>
<weights .../>
</layer>

ReLU Layer

Back to top

Name: ReLU

Category: Activation

Short description: Reference

Detailed description: Reference

Parameters: ReLU layer parameters are specified parameters in the data node, which is a child of the layer node.

  • Parameter name: negative_slope
    • Description: negative_slope is a multiplier, which is used if the unit is not active (that is, negative). For example, negative_slope equal to 0.1 means that an inactive unit value would be multiplied by 0.1 and this is the Leaky ReLU. If negative_slope is equal to 0.0, this is the usual ReLU.
    • Range of values: a non-negative floating-point number
    • Type: float
    • Default value: None
    • Required: no

Mathematical Formulation

\[ Y_{i}^{( l )} = max(0, Y_{i}^{( l - 1 )}) \]

Inputs:

  • 1: Multidimensional input blob. Required.

Example

<layer ... type="ReLU" ... >
<data negative_slope="0.100000"/>
<input> ... </input>
<output> ... </output>
</layer>

ReorgYolo Layer

Back to top

Name: ReorgYolo

Category: Layer

Short description: ReorgYolo reorganizes input blob taking into account strides.

Detailed description: Reference

Parameters: ReorgYolo layer parameters are specified parameters in the data node, which is a child of the layer node.

  • Parameter name: stride
    • Description: stride is the distance between cut throws in output blobs.
    • Range of values: an integer
    • Type: int
    • Default value: None
    • Required: yes

Inputs:

  • 1: 4D input blob. Required.

Example

<layer ... type="ReorgYolo" ... >
<data stride="1"/>
<input> ... </input>
<output> ... </output>
</layer>

Resample (Type 1) Layer

Back to top

Name: Resample

Category: Layer

Short description: Resample layer scales the input blob by the specified parameters.

Parameters: Resample layer parameters are specified in the data node, which is a child of the layer node. Resample Type 1 layer has one input blob containing image to resample.

  • Parameter name: type
    • Description: type parameter specifies the type of blob interpolation.
    • Range of values:
      • caffe.ResampleParameter.LINEAR - linear blob interpolation
      • caffe.ResampleParameter.NEAREST - nearest-neighbor blob interpolation
    • Type: string
    • Default value: None
    • Required: yes
  • Parameter name: antialias
    • Description: antialias is a flag that specifies whether to perform anti-aliasing.
    • Range of values:
      • 0 - do not perform anti-aliasing
      • 1 - perform anti-aliasing
    • Type: int
    • Default value: 0
    • Required: no
  • Parameter name: factor
    • Description: factor specifies a scale factor for output height and width.
    • Range of values: a positive integer
    • Type: int
    • Default value: None
    • Required: yes

Inputs:

  • 1: 4D input blob. Required.

Example

<layer type="Resample">
<data antialias="0" factor="2" type="caffe.ResampleParameter.LINEAR"/>
<input>
<port id="0">
<dim>1</dim>
<dim>3</dim>
<dim>25</dim>
<dim>30</dim>
</port>
</input>
<output>
<port id="1">
<dim>1</dim>
<dim>3</dim>
<dim>50</dim>
<dim>60</dim>
</port>
</output>
</layer>

Resample (Type 2) Layer

Back to top

Name: Resample

Category: Layer

Short description: Resample layer scales the input blob by the specified parameters.

Parameters: Resample layer parameters are specified in the data node, which is a child of the layer node. Resample Type 2 layer has two input blobs containing image to resample and output dimensions.

  • Parameter name: type
    • Description: type parameter specifies the type of blob interpolation.
    • Range of values:
      • caffe.ResampleParameter.LINEAR - linear blob interpolation
      • caffe.ResampleParameter.NEAREST - nearest-neighbor blob interpolation
    • Type: string
    • Default value: None
    • Required: yes
  • Parameter name: antialias
    • Description: antialias is a flag that specifies whether to perform anti-aliasing.
    • Range of values:
      • 0 - do not perform anti-aliasing
      • 1 - perform anti-aliasing
    • Type: int
    • Default value: 0
    • Required: no
  • Parameter name: factor
    • Description: factor parameter is ignored in the Resample Type 2.
    • Range of values: a positive integer
    • Type: int
    • Default value: None
    • Required: yes

Inputs:

  • 1: 4D input blob. Required.
  • 2: 1D blob describing output shape. Required.

Example

<layer type="Resample">
<data antialias="0" factor="1" type="caffe.ResampleParameter.LINEAR"/>
<input>
<port id="0">
<dim>1</dim>
<dim>3</dim>
<dim>25</dim>
<dim>30</dim>
</port>
<port id="1">
<dim>4</dim> <!--The values in this input are [1,3,50,60] -->
</port>
</input>
<output>
<port id="2">
<dim>1</dim>
<dim>3</dim>
<dim>50</dim>
<dim>60</dim>
</port>
</output>
</layer>

Reshape Layer

Back to top

Name: Reshape

Category: Layer

Short description: Reshape layer changes dimensions of the input blob according to the specified order. Input blob volume is equal to output blob volume, where volume is the product of dimensions.

Detailed description: Reference

Parameters: Reshape layer does not have parameters. Reshape layer takes two input blobs: the blob to be resized and the output blob shape. The values in the second blob can be -1, 0 and any positive integer number. The two special values -1 and 0:

  • 0 means copying the respective dimension of the input blob.
  • -1 means that this dimension is calculated to keep the overall elements count the same as in the input blob. No more than one -1 can be used in a reshape operation.

Inputs:

  • 1: Multidimensional input blob. Required.
  • 2: 1D blob describing output shape. Required.

Example

<layer ... type="Reshape" ...>
<input>
<port id="0">
<dim>2</dim>
<dim>5</dim>
<dim>5</dim>
<dim>24</dim>
</port>
<port id="1">
<dim>3</dim> <!--The blob contains 3 elements: 0, -1, 4 -->
</port>
</input>
<output>
<port id="2">
<dim>2</dim>
<dim>150</dim>
<dim>4</dim>
</port>
</output>
</layer>

ReverseSequence Layer

Back to top

Name: ReverseSequence

Category: Layer

Short description: ReverseSequence reverses variable length slices of data.

Detailed description: ReverseSequence slices input along the dimension specified in the batch_axis, and for each slice i, reverses the first lengths[i] (the second input) elements along the dimension specified in the seq_axis.

Parameters: ReverseSequence layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: batch_axis
    • Description: batch_axis is the index of the batch dimension.
    • Range of values: an integer. Can be negative.
    • Type: int
    • Default value: 0
    • Required: no
  • Parameter name: seq_axis
    • Description: seq_axis is the index of the sequence dimension.
    • Range of values: an integer. Can be negative.
    • Type: int
    • Default value: 1
    • Required: no

Inputs:

  • 1: Blob with input data to reverse. Required.
  • 2: 1D blob with sequence lengths in the first input blob. Required.

Example

<layer ... type="ReverseSequence">
<data batch_axis="0" seq_axis="1"/>
<input>
<port id="0">
<dim>3</dim>
<dim>10</dim>
<dim>100</dim>
<dim>200</dim>
</port>
<port id="1">
<dim>10</dim>
</port>
</input>
<output>
<port id="2">
<dim>3</dim>
<dim>10</dim>
<dim>100</dim>
<dim>200</dim>
</port>
</output>
</layer>

RNNCell Layer

Back to top

Name: RNNCell

Category: Layer

Short description: RNNCell layer computes the output using the formula described in the article.

Parameters: RNNCell layer parameters should be specified as the data node, which is a child of the layer node.

  • Parameter name: hidden_size
    • Description: hidden_size specifies hidden state size.
    • Range of values: a positive integer
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: activations
    • Description: activation functions for gates
    • Range of values: any combination of relu, sigmoid, tanh
    • Type: a list of strings
    • Default value: sigmoid,tanh
    • Required: no
  • Parameter name: activations_alpha, activations_beta
    • Description: activations_alpha, activations_beta functions parameters
    • Range of values: a list of floating-point numbers
    • Type: float[]
    • Default value: None
    • Required: no
  • Parameter name: clip
    • Description: clip specifies value for tensor clipping to be in [-C, C] before activations
    • Range of values: a positive floating-point number
    • Type: float
    • Default value: None
    • Required: no

Inputs

  • 1: X - 2D ([batch_size, input_size]) input data. Required.
  • 2: Hi - 2D ([batch_size, hidden_size]) input hidden state data. Required.

Outputs

  • 1: Ho - 2D ([batch_size, hidden_size]) output hidden state.

ROIPooling Layer

Back to top

Name: ROIPooling

Category: Pool

Short description: ROIPooling is a pooling layer used over feature maps of non-uniform input sizes and outputs a feature map of a fixed size.

Detailed description: deepsense.io reference

Parameters: ROIPooling layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: pooled_h
    • Description: pooled_h is the height of the ROI output feature map. For example, pooled_h equal to 6 means that the height of the output of ROIPooling is 6.
    • Range of values: a non-negavive integer
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: pooled_w
    • Description: pooled_w is the width of the ROI output feature map. For example, pooled_w equal to 6 means that the width of the output of ROIPooling is 6.
    • Range of values: a non-negative integer
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: spatial_scale
    • Description: spatial_scale is the ratio of the input feature map over the input image size.
    • Range of values: a positive floating-point number
    • Type: float
    • Default value: None
    • Required: yes
  • Parameter name: method
    • Description: method specifies a method to perform pooling. If the method is bilinear, the input box coordinates are normalized to the [0,1] interval.
    • Range of values: max or bilinear
    • Type: string
    • Default value: max
    • Required: no

Inputs:

  • 1: 4D input blob with feature maps. Required.
  • 2: 2D input blob describing box consisting of 5 element tuples: [batch_id, x_1, y_1, x_2, y_2]. Required.

Mathematical Formulation

\[ output_{j} = MAX\{ x_{0}, ... x_{i}\} \]

Example

<layer ... type="ROIPooling" ... >
<data pooled_h="6" pooled_w="6" spatial_scale="0.062500"/>
<input> ... </input>
<output> ... </output>
</layer>

ExperimentalDetectronROIFeatureExtractor Layer

Back to top

Name: ExperimentalDetectronROIFeatureExtractor (aka ROIAlignPyramid)

Category: Pool

Short description: ExperimentalDetectronROIFeatureExtractor is the ROIAlign operation applied over a feature pyramid.

Detailed description: ExperimentalDetectronROIFeatureExtractor maps input ROIs to the levels of the pyramid depending on the sizes of ROIs and parameters of the operation, and then extracts features via ROIAlign from corresponding pyramid levels. For more details please see the math formulas below and the following sources:

Parameters: ExperimentalDetectronROIFeatureExtractor layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: output_size
    • Description: output_size is the width and height of the output tensor.
    • Range of values: a positive integer number
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: sampling_ratio
    • Description: sampling_ratio is the number of sampling points per the output value. If 0, then use adaptive number computed as ceil(roi_width / output_width), and likewise for height.
    • Range of values: a non-negative integer number
    • Type: int
    • Default value: 0
    • Required: yes
  • Parameter name: pyramid_scales
    • Description: pyramid_scales enlists image_size / layer_size[l] ratios for pyramid layers l=1,...,L, where L is the number of pyramid layers, and image_size refers to network's input image. Note that pyramid's largest layer may have smaller size than input image, e.g. image_size is 640 in the XML example below.
    • Range of values: a list of positive integer numbers
    • Type: int[]
    • Default value: None
    • Required: yes

Inputs:

  • 0: 2D input blob describing the rois as 4-tuples: [x1, y1, x2, y2]. Batch size is the number of rois. Coordinates x and y are float numbers and refer to the input image_size. Required.
  • 1, ..., L: Pyramid of 4D input blobs with feature maps. Batch size must be 1. The number of channels must be the same for all layers of the pyramid. The layer width and height must equal to the layer_size[l] = image_size / pyramid_scales[l]. Required.

Outputs:

  • 0: 4D output blob. Batch size equals to number of rois. Channels number is the same as for all images in the input pyramid. Data type is float. Required.

Mathematical Formulation

ExperimentalDetectronROIFeatureExtractor applies the ROIAlign algorithm to the pyramid layers:

  • output[i, :, :, :] = ROIAlign(inputPyramid[j], rois[i])
  • j = PyramidLevelMapper(rois[i])

PyramidLevelMapper maps the ROI to the pyramid level using the following formula:

  • j = floor(2 + log2(sqrt(w * h) / 224)

Example

<layer ... type="ExperimentalDetectronROIFeatureExtractor">
<data output_size="14"
pyramid_scales="4,8,16,32"
sampling_ratio="2"/>
<input>
<port id="0">
<dim>100</dim>
<dim>4</dim>
</port>
<port id="1">
<dim>1</dim>
<dim>256</dim>
<dim>160</dim>
<dim>160</dim>
</port>
<port id="2">
<dim>1</dim>
<dim>256</dim>
<dim>80</dim>
<dim>80</dim>
</port>
<port id="3">
<dim>1</dim>
<dim>256</dim>
<dim>40</dim>
<dim>40</dim>
</port>
<port id="4">
<dim>1</dim>
<dim>256</dim>
<dim>20</dim>
<dim>20</dim>
</port>
</input>
<output>
<port id="5">
<dim>100</dim>
<dim>256</dim>
<dim>14</dim>
<dim>14</dim>
</port>
</output>
</layer>

ExperimentalSparseWeightedSum Layer

Back to top

Name: ExperimentalSparseWeightedSum

Category: Layer

Short description: ExperimentalSparseWeightedSum extracts embedding vectors from the parameters table for each object feature value and sum up these embedding vectors multiplied by weights for each object.

Detailed description: Reference. This is similar to embedding_lookup_sparse but it accepts objects with empty feature values for which it uses a default value to extract an embedding from the parameters table. In comparison with embedding_lookup_sparse it has a limitation to work only with two-dimensional indices tensor.

Inputs:

  • 1: 2-D tensor. Input indices of the sparse tensor. It contains with an integer type. Required.
  • 2: 1-D tensor. Input values of the sparse tensor. It contains with an integer type. Required.
  • 3: 1-D tensor. Dense shape of the sparse tensor. It contains with an integer type. Required.
  • 4: N-D tensor. The parameters table. It contains with a float type. Required.
  • 5: 0-D tensor. The default value. It contains with an integer type. Required.
  • 6: 1-D tensor. Input weights. It contains with a float type. Optional.

Outputs:

  • 1: The output tensor of resulted embedding vectors for each object. It is has a shape [batch_size, params_table_shape[1], ..., params_table_shape[-1]] where batch_size is a number of objects or a number of rows in the sparse tensor.

ScaleShift Layer

Back to top

Name: ScaleShift

Category: Layer

Short description: ScaleShift layer performs linear transformation of the input blobs. Weights denote a scaling parameter, biases denote a shift.

Parameters: ScaleShift layer does not have parameters.

Inputs:

  • 1: 4D input blob. Required.

Mathematical Formulation

\[ o_{i} =\gamma b_{i} + \beta \]

Example

<layer ... type="ScaleShift" ... >
<input> ... </input>
<output> ... </output>
</layer>

Select Layer

Back to top Name: Select

Category: Layer

Short description: Select layer returns a tensor filled with the elements from the second or the third input, depending on the condition (the first input) value.

Detailed description: Select takes elements from the second (then) or the third (else) input based on a condition mask provided in the first input (cond). The cond tensor is broadcasted to then and else tensors. The output tensor shape is equal to the broadcasted shape of cond, then, and else. The behavior is similar to numpy.where with three parameters.

Parameters: Select layer does not have parameters.

Inputs:

  • 1: cond tensor with selection mask (only integer values). The tensor can be 0D.
  • 2: then the tensor with elements to take where condition is true.
  • 3: else the tensor with elements to take where condition is false.

Example

<layer ... type="Select">
<input>
<port id="0"> <!-- cond value is: [[0, 0], [1, 0], [1, 1]]-->
<dim>3</dim>
<dim>2</dim>
</port>
<port id="1"> <!-- then value is: [[-1, 0], [1, 2], [3, 4]]-->
<dim>3</dim>
<dim>2</dim>
</port>
<port id="2"> <!-- else value is: [[11, 10], [9, 8], [7, 6]]-->
<dim>3</dim>
<dim>2</dim>
</port>
</input>
<output>
<port id="1"> <!-- output value is: [[11, 10], [1, 8], [3, 4]]-->
<dim>3</dim>
<dim>2</dim>
</port>
</output>
</layer>

Shape Layer

Back to top

Name: Shape

Category: Layer

Short description: Shape produces a blob with the input blob shape.

Parameters: Shape layer does not have parameters.

Inputs:

  • 1: Multidimensional input blob. Required.

Example

<layer ... type="Shape">
<input>
<port id="0">
<dim>2</dim>
<dim>3</dim>
<dim>224</dim>
<dim>224</dim>
</port>
</input>
<output>
<port id="1"> <!-- output value is: [2,3,224,224]-->
<dim>4</dim>
</port>
</output>
</layer>

ShuffleChannels Layer

Back to top

Name: ShuffleChannels

Category: Layer

Short description: ShuffleChannels permutes data in the channel dimension of the input blob.

Parameters: ShuffleChannels layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: axis
    • Description: axis specifies the index of a channel dimension.
    • Range of values: a non-negative integer
    • Type: int
    • Default value: 1
    • Required: No
  • Parameter name: group
    • Description: group specifies the number of groups to split the channel dimension into. This number must evenly divide the channel dimension size.
    • Range of values: a positive integer
    • Type: int
    • Default value: 1
    • Required: No

Inputs:

  • 1: 4D input blob. Required.

Mathematical Formulation

The operation is the equivalent with the following transformation of the input blob x of shape [N, C, H, W]:

x' = reshape(x, [N, group, C / group, H * W])
x'' = transpose(x', [0, 2, 1, 3])
y = reshape(x'', [N, C, H, W])

where group is the layer parameter described above.

Example

<layer ... type="ShuffleChannels" ...>
<data group="3" axis="1"/>
<input>
<port id="0">
<dim>3</dim>
<dim>12</dim>
<dim>200</dim>
<dim>400</dim>
</port>
</input>
<output>
<port id="1">
<dim>3</dim>
<dim>12</dim>
<dim>200</dim>
<dim>400</dim>
</port>
</output>
</layer>

SimplerNMS Layer

Back to top

Name: SimplerNMS

Category: Layer

Short description: SimplerNMS layer filters bounding boxes and outputs only those with the highest confidence of prediction.

Parameters: SimplerNMS layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: pre_nms_topn
    • Description: pre_nms_topn is the number of bounding boxes before the NMS operation. For example, pre_nms_topn equal to 15 means that the minimum box size is 15.
    • Range of values: a positive integer
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: post_nms_topn
    • Description: post_nms_topn is the quantity of bounding boxes after the NMS operation. For example, post_nms_topn equal to 15 means that the maximum box size is 15.
    • Range of values: a positive integer
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: iou_threshold
    • Description: iou_threshold is the minimum ratio of boxes overlapping to be taken into consideration. For example, iou_threshold equal to 0.7 means that all boxes with overlapping ratio less than 0.7 are filtered out.
    • Range of values: a positive floating-point number
    • Type: float
    • Default value: None
    • Required: yes
  • Parameter name: feat_stride
    • Description: feat_stride is the step size to slide over boxes (in pixels). For example, feat_stride equal to 16 means that all boxes are analyzed with the slide 16.
    • Range of values: a positive integer
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: min_bbox_size
    • Description: min_bbox_size is the minimum size of a box to be taken into consideration.
    • Range of values: a positive integer
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: scale
    • Description: scale is for generating anchor boxes.
    • Range of values: a list of positive floating-point numbers
    • Type: float[]
    • Default value: []
    • Required: no

Inputs:

  • 1: 4D input blob with class prediction scores. Required.
  • 2: 4D input blob with box logits. Required.
  • 3: 1D input blob 3 or 4 elements: [image height, image width, scale for image height/width OR scale for image height and scale for image width]. Required.

Mathematical Formulation

SimplerNMS accepts three inputs with four dimensions. Produced blob has two dimensions, the first one equals post_nms_topn. SimplerNMS does the following with the input blob:

  1. Generates initial anchor boxes. Left top corner of all boxes is (0, 0). Width and height of boxes are calculated based on scaled (according to the scale parameter) default widths and heights
  2. For each point in the first input blob:
    • pins anchor boxes to a picture according to the second input blob, which contains four deltas for each box: for x and y of the center, for width, and for height
    • finds out score in the first input blob
  3. Filters out boxes with size less than min_bbox_size.
  4. Sorts all proposals (box, score) by score from highest to lowest
  5. Takes top pre_nms_topn proposals
  6. Calculates intersections for boxes and filters out all with \(intersection/union > iou\_threshold\)
  7. Takes top post_nms_topn proposals
  8. Returns top proposals

Example

<layer ... type="SimplerNMS" ... >
<data iou_threshold="0.700000" min_bbox_size="16" feat_stride="16" pre_nms_topn="6000" post_nms_topn="150"/>
<input> ... </input>
<output> ... </output>
</layer>

Slice Layer

Back to top

Name: Slice

Category: Layer

Short description: Slice layer splits the input blob into several pieces over the specified axis.

Parameters: Slice layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: axis
    • Description: axis specifies the axis to split the input blob along
    • Range of values: a non-negative integer
    • Type: int
    • Default value: None
    • Required: yes

Inputs:

  • 1: Multidimensional input blob. Required.

Example

<layer ... type="Slice" ...>
<data axis="1"/>
<input>
<port id="0">
<dim>1</dim>
<dim>1048</dim>
<dim>14</dim>
<dim>14</dim>
</port>
</input>
<output>
<port id="1">
<dim>1</dim>
<dim>1024</dim>
<dim>14</dim>
<dim>14</dim>
</port>
<port id="2">
<dim>1</dim>
<dim>24</dim>
<dim>14</dim>
<dim>14</dim>
</port>
</output>
</layer>

SoftMax Layer

Back to top

Name: SoftMax

Category: Activation

Short description: Reference

Detailed description: Reference

Parameters: SoftMax layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: axis
    • Description: axis is the axis along which the SoftMax is calculated.
    • Range of values: a positive integer
    • Type: int
    • Default value: 1
    • Required: no

Mathematical Formulation

\[ y_{c} = \frac{e^{Z_{c}}}{\sum_{d=1}^{C}e^{Z_{d}}} \]

where \(C\) is a number of classes

Example

<layer ... type="SoftMax" ... >
<data axis="1" />
<input> ... </input>
<output> ... </output>
</layer>

Inputs:

  • 1: Multidimensional input blob. Required.

SparseFillEmptyRows Layer

Back to top

Name: SparseFillEmptyRows

Category: Layer

Short description: SparseFillEmptyRows fills empty rows in the input 2-D SparseTensor with a default value.

Detailed description: Reference

Inputs:

  • 1: 2-D tensor. Input indices of the sparse tensor. Required.
  • 2: 1-D tensor. Input values of the sparse tensor. Required.
  • 3: 1-D tensor. Shape of the sparse tensor. Value of this input is required for the Model Optimizer.
  • 4: 0-D tensor. Default value to insert at rows missing from the input sparse tensor. Required.

Outputs:

  • 1: 2-D tensor. Indices of the filled sparse tensor.
  • 2: 1-D tensor. Values of the filled sparse tensor.
  • 3: 1-D tensor. An indicator of whether the dense row was missing in the input sparse tensor.

SparseSegmentMean Layer

Back to top

Name: SparseSegmentMean

Category: Layer

Short description: SparseSegmentMean computes the mean along sparse segments of a tensor.

Detailed description: Reference

Parameters: SparseSegmentMean layer does not have parameters.

Inputs:

  • 1: ND tensor. Data tensor from which rows are selected for the mean operation. Required.
  • 2: 1D tensor. Tensor of rows indices selected from the first input tensor along 0 dimension. Required.
  • 3: 1D tensor. Tensor of segment IDs that rows selected for the operation belong to. Rows belonging to the same segment are summed up and divided by N, where N is a number of selected rows in a segment. This input has the same size as the second input. Values must be sorted in ascending order and can be repeated. Required.

Outputs:

  • 1: ND tensor. It has the same shape as the data tensor, except for dimension 0, which has a size equal to a size of an indices tensor.

SparseSegmentSqrtN Layer

Back to top

Name: SparseSegmentSqrtN

Category: Layer

Short description: SparseSegmentSqrtN computes the sum along sparse segments of a tensor and divides it by the square root of N, where N is a number of rows in a segment.

Detailed description: Reference

Parameters: SparseSegmentSqrtN layer does not have parameters.

Inputs:

  • 1: ND tensor. Data tensor from which rows are selected. Required.
  • 2: 1D tensor. Tensor of rows indices selected from the first input tensor along 0 dimension. Required.
  • 3: 1D tensor. Tensor of segment IDs that selected rows belong to. Rows belonging to the same segment are summed up and divided by the square root of N, where N is a number of rows in a segment. This input tensor has the same size as the second input. Values must be sorted in ascending order and can be repeated. Required.

Outputs:

  • 1: ND tensor. It has the same shape as the data tensor, except for a dimension 0, which has a size equal to a size of an indices tensor.

SparseSegmentSum Layer

Back to top

Name: SparseSegmentSum

Category: Layer

Short description: SparseSegmentSum computes the sum along sparse segments of a tensor.

Detailed description: Reference

Parameters: SparseSegmentSum layer does not have parameters.

Inputs:

  • 1: ND tensor. Data tensor from which rows are selected. Required.
  • 2: 1D tensor. Tensor of rows indices selected from the first input tensor along 0 dimension. Required.
  • 3: 1D tensor. Tensor of segment IDs that selected rows belong to. Rows belonging to the same segment are summed up. This input tensor has the same size as the second input. Values must be sorted in ascending order and can be repeated. Required.

Outputs:

  • 1: ND tensor. It has the same shape as the data tensor, except for a dimension 0, which has a size equal to a size of an indices tensor.

SparseToDense Layer

Back to top

Name: SparseToDense

Category: Layer

Short description: SparseToDense converts a sparse tensor into a dense tensor.

Detailed description: Reference

Inputs:

  • 1: 2-D tensor. Input indices of the sparse tensor. It contains with an integer type. Required.
  • 2: 1-D tensor. Dense shape of the sparse tensor. It contains with an integer type. Required.
  • 3: 1-D tensor. Input values of the sparse tensor. It contains with integer and float types. Required.
  • 4: 0-D tensor. Default value to insert at missing positions. The fourth input type must be the same as the third input type. If it is not specified, zero value is used. Optional.

Outputs:

  • 1: The output dense tensor. The output tensor shape is equal to a value of the second input.

Split Layer

Back to top

Name: Split

Category: Layer

Short description: Split layer splits the input along the specified axis into several output pieces.

Detailed description: Reference

Parameters: Split layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: axis
    • Description: axis is the number of the axis to split input blob along.
    • Range of values: a non-negative integer less than the number of dimensions in the input
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: num_split
    • Description: num_split is the number of pieces to split the input into. The num_split must evenly divide the size of the axis dimension.
    • Range of values: a positive integer less than or equal to the size of the dimension being split over
    • Type: int
    • Default value: None
    • Required: yes

Mathematical Formulation

For example, if the blob is BxC+CxHxW, axis="1", and num_split="2", the sizes of output blobs are BxCxHxW.

Inputs:

  • 1: Multidimensional input blob. Required.

Example

<layer ... type="Split" ... >
<data axis="0" num_split="2"/>
<input> ... </input>
<output> ... </output>
</layer>

Squeeze Layer

Name: Squeeze

Category: Layer

Short description: Squeeze removes specified dimensions (second input) equal to 1 of the first input tensor. If the second input is omitted then all dimensions equal to 1 are removed. If the specified dimension is not equal to one then error is raised.

Parameters: Squeeze layer doesn't have parameters.

Inputs:

  • 1: Multidimensional input blob. Required.
  • 2: (optional): 0D or 1D tensor with dimensions indices to squeeze. Values could be negative. Indices could be integer or float values.

Example

Example 1:

<layer type="Squeeze">
<input>
<port id="0">
<dim>1</dim>
<dim>3</dim>
<dim>1</dim>
<dim>2</dim>
</port>
</input>
<input>
<port id="1">
<dim>2</dim> <!-- value [0, 2] -->
</port>
</input>
<output>
<port id="2">
<dim>3</dim>
<dim>2</dim>
</port>
</output>
</layer>

Example 2: squeeze 1D tensor with 1 element to a 0D tensor (constant)

<layer type="Squeeze">
<input>
<port id="0">
<dim>1</dim>
</port>
</input>
<input>
<port id="1">
<dim>1</dim> <!-- value is [0] -->
</port>
</input>
<output>
<port id="2">
</port>
</output>
</layer>

StridedSlice Layer

Name: StridedSlice

Short description: StridedSlice layer extracts a strided slice of a blob. It is similar to generalized array indexing in Python*.

Parameters: StridedSlice layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: begin_mask
    • Description: begin_mask is a bit mask. begin_mask[i] equal to 0 means that the corresponding dimension of the begin input is ignored.
    • Range of values: a list of 0s and 1s
    • Type: int[]
    • Default value: [1]
    • Required: yes
  • Parameter name: end_mask
    • Description: end_mask is a bit mask. If end_mask[i] is 0, the corresponding dimension of the end input is ignored.
    • Range of values: a list of 0s and 1s
    • Type: int[]
    • Default value: [1]
    • Required: yes
  • Parameter name: new_axis_mask
    • Description: new_axis_mask is a bit mask. If new_axis_mask[i] is 1, a length 1 dimension is inserted on the i-th position of input blob.
    • Range of values: a list of 0s and 1s
    • Type: int[]
    • Default value: [0]
    • Required: no
  • Parameter name: shrink_axis_mask
    • Description: shrink_axis_mask is a bit mask. If shrink_axis_mask[i] is 1, the dimension on the i-th position is deleted.
    • Range of values: a list of 0s and 1s
    • Type: int[]
    • Default value: [0]
    • Required: no
  • Parameter name: ellipsis_mask
    • Description: ellipsis_mask is a bit mask. It inserts missing dimensions on a position of a non-zero bit.
    • Range of values: a list of 0s and 1. Only one non-zero bit is allowed.
    • Type: int[]
    • Default value: [0]
    • Required: no

Inputs:

  • 1: Multidimensional input blob. Required.
  • 2: begin input - 1D input blob with begin indexes for input blob slicing. Required. Out-of-bounds values are silently clamped. If begin_mask[i] is 0, the value of begin[i] is ignored and the range of the appropriate dimension starts from 0. Negative values mean indexing starts from the end. For example, if foo=[1,2,3], begin[0]=-1 means begin[0]=3.
  • 3: end input - 1D input blob with end indexes for input blob slicing. Required. Out-of-bounds values will be silently clamped. If end_mask[i] is 0, the value of end[i] is ignored and the full range of the appropriate dimension is used instead. Negative values mean indexing starts from the end. For example, if foo=[1,2,3], end[0]=-1 means end[0]=3.
  • 4: stride input - 1D input blob with strides. Optional.

Example

<layer ... type="StridedSlice" ...>
<data begin_mask="0,1,0,0,0" ellipsis_mask="0,0,0,0,0" end_mask="0,1,0,0,0" new_axis_mask="0,0,0,0,0" shrink_axis_mask="0,1,0,0,0"/>
<input>
<port id="0">
<dim>1</dim>
<dim>2</dim>
<dim>384</dim>
<dim>640</dim>
<dim>8</dim>
</port>
<port id="1">
<dim>5</dim>
</port>
<port id="2">
<dim>5</dim>
</port>
<port id="3">
<dim>5</dim>
</port>
</input>
<output>
<port id="4">
<dim>1</dim>
<dim>384</dim>
<dim>640</dim>
<dim>8</dim>
</port>
</output>
</layer>

TensorIterator Layer

Back to top

Name: TensorIterator

Category: Layer

Short description: TensorIterator (TI) layer performs recurrent sub-graph execution iterating through the data.

Parameters: The parameters are specified in the child nodes of the port_map and back_edges sections, which are child nodes of the layer node. The port_map and back_edges sections specify data mapping rules.

  • Node: port_map is a set of rules to map input/output data blobs of the TensorIterator layer onto body data blobs. Port mapping rule is presented as input/output nodes.
    • Parameter name: external_port_id
      • Description: external_port_id is a port ID of the TensorIterator layer.
      • Range of values: indexes of the TensorIterator outputs
      • Type: int
      • Default value: None
      • Required: yes
    • Parameter name: internal_layer_id
      • Description: internal_layer_id is a layer ID inside the body sub-network to map to.
      • Range of values: IDs of the layers inside in the TensorIterator layer
      • Type: int
      • Default value: None
      • Required: yes
    • Parameter name: internal_port_id
      • Description: internal_port_id is a port ID of the body layer to map to.
      • Range of values: indexes of the body layer input
      • Type: int
      • Default value: None
      • Required: yes
    • Parameter name: axis
      • Description: axis is an axis to iterate through. -1 means no iteration is done.
      • Range of values: an integer
      • Type: int
      • Default value: -1
      • Required: no
    • Parameter name: start
      • Description: start is an index where the iteration starts from. Negative value means counting indexes from the end.
      • Range of values: an integer
      • Type: int
      • Default value: 0
      • Required: no
    • Parameter name: end
      • Description: end is an index where iteration ends. Negative value means counting indexes from the end.
      • Range of values: an integer
      • Type: int
      • Default value: -1
      • Required: no
    • Parameter name: stride
      • Description: stride is a step of iteration. Negative value means backward iteration.
      • Range of values: an integer
      • Type: int
      • Default value: 1
      • Required: no
  • Node: back_edges is a set of rules to transfer data blobs between body iteration. Mapping rule is presented as a general edge node with port and layer indexes of body sub-network.
    • Parameter name: from-layer
      • Description: from-layer is a layer ID inside the body sub-network.
      • Range of values: IDs of the layers inside the TensorIterator
      • Type: int
      • Default value: None
      • Required: yes
    • Parameter name: from-port
      • Description: from-port is a port ID inside the body sub-network to start mapping from.
      • Range of values: the respective layer port index
      • Type: int
      • Default value: None
      • Required: yes
    • Parameter name: to-layer
      • Description: to-layer is a layer ID inside the body sub-network to end mapping.
      • Range of values: IDs of the layers inside the TensorIterator
      • Type: int
      • Default value: None
      • Required: yes
    • Parameter name: to-port
      • Description: to-port is a port ID inside the body sub-network to end mapping.
      • Range of values: the respective layer port index
      • Type: int
      • Default value: None
      • Required: yes
  • Node: body is a sub-network that will be recurrently executed.
    • Parameters: The body node does not have parameters.

Example

<layer ... type="TensorIterator" ... >
<input> ... </input>
<output> ... </output>
<port_map>
<input external_port_id="0" internal_layer_id="0" internal_port_id="0" axis="1" start="-1" end="0" stride="-1"/>
<input external_port_id="1" internal_layer_id="1" internal_port_id="1"/>
...
<output external_port_id="3" internal_layer_id="2" internal_port_id="1" axis="1" start="-1" end="0" stride="-1"/>
...
</port_map>
<back_edges>
<edge from-layer="1" from-port="1" to-layer="1" to-port="1"/>
...
</back_edges>
<body>
<layers> ... </layers>
<edges> ... </edges>
</body>
</layer>

Tile Layer

Back to top

Name: Tile

Category: Layer

Short description: Tile layer extends input blob with copies of data along a specified axis.

Detailed description: Reference

Parameters: Tile layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: axis
    • Description: axis is the index of an axis to tile. For example, axis equal to 3 means that the fourth axis is used for tiling.
    • Range of values: a positive integer
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: tiles
    • Description: tiles is the size of the specified axis in the output blob. For example, tiles equal to 88 means that the output blob gets 88 copies of data from the specified axis.
    • Range of values: a positive integer
    • Type: int
    • Default value: None
    • Required: yes

Mathematical Formulation

Tile extends input blobs and filling in output blobs by the following rules:

\[ out_i=input_i[inner\_dim*t] \]

\[ t \in \left ( 0, \quad tiles \right ) \]

Inputs:

  • 1: Multidimensional input blob. Required.

Example

<layer ... type="Tile" ... >
<data axis="3" tiles="88"/>
<input> ... </input>
<output> ... </output>
</layer>

TopK Layer

Back to top

Name: TopK

Category: Layer

Short description: TopK layer computes indices and values of the k maximum/minimum values for each slice along the axis specified.

Parameters: TopK layer parameters are specified in the data node, which is a child of the layer node.

  • Parameter name: axis
    • Description: Specifies the axis along which to search for k maximum/minimum values.
    • Range of values: an integer. Negative value means counting dimension from the end.
    • Type: int
    • Default value: None
    • Required: yes
  • Parameter name: mode
    • Description: mode specifies an operation to use for selecting the largest element of two.
    • Range of values: min, max
    • Type: string
    • Default value: None
    • Required: yes
  • Parameter name: sort
    • Description: sort specifies an order of output elements and/or indices.
    • Range of values: value, index, none
    • Type: string
    • Default value: None
    • Required: yes

Inputs:

  • 1: Arbitrary tensor. Required.
  • 2: k - scalar specifies how many maximum/minimum elements should be computed

Outputs:

  • 1: Output tensor with top k values from input tensor along specified dimension axis. The shape of the tensor is [input1.shape[0], ..., input1.shape[axis-1], k, input1.shape[axis+1], ...].
  • 2: Output tensor with top k indices for each slice along axis dimension. The shape of the tensor is the same as for the 1st output, that is [input1.shape[0], ..., input1.shape[axis-1], k, input1.shape[axis+1], ...]

Mathematical Formulation

Output tensor is populated by values computes in the following way:

\[ output[i1, ..., i(axis-1), j, i(axis+1) ..., iN] = top_k(input[i1, ...., i(axis-1), :, i(axis+1), ..., iN]), k, sort, mode) \]

So for each slice input[i1, ...., i(axis-1), :, i(axis+1), ..., iN] which represents 1D array, top_k value is computed individually. Sorting and minimum/maximum are controlled by sort and mode attributes.

Example

<layer ... type="TopK" ... >
<data axis="1" mode="max" sort="value"/>
<input>
<port id="0">
<dim>6</dim>
<dim>12</dim>
<dim>10</dim>
<dim>24</dim>
</port>
<port id="1"> <!-- value k = 3 -->
</port>
<output>
<port id="2">
<dim>6</dim>
<dim>3</dim>
<dim>10</dim>
<dim>24</dim>
</port>
</output>
</layer>

Unique Layer

Back to top

Name: Unique

Category: Layer

Short description: Unique finds unique elements in 1-D tensor.

Detailed description: Reference

Parameters: Unique layer parameters should be specified in the data node, which is a child of the layer node.

  • Parameter name: sorted
    • Description: If sorted is equal to true, the unique elements in the output are sorted in ascending order. Otherwise, all of the unique elements are sorted in the same order as they occur in the input.
    • Range of values: true or false
    • Type: string
    • Required: yes
  • Parameter name: return_inverse
    • Description: If return_inverse is equal to true, the layer outputs the indices. Otherwise, it does not.
    • Range of values: true or false
    • Type: string
    • Required: yes
  • Parameter name: return_counts
    • Description: If return_counts is equal to true, the layer outputs the counts for each unique element. Otherwise, it does not.
    • Range of values: true or false
    • Type: string
    • Required: yes

Input:

  • 1: 1-D tensor. Input tensor. Required.

Outputs:

  • 1: 1-D tensor. Tensor of all unique elements from the input tensor. As a number of unique elements can be less than a size of the input, the end of this tensor is marked with the latest unique element. Required.
  • 2: 1-D tensor. Tensor of indices of unique elements of the first output that can be used to reconstruct the input. The size of this tensor is equal to the input size. It outputs in the second output port. Optional.
  • 3: 1-D tensor. Tensor of counts of occurrences of each unique element in the input. It has the same size as the output with unique elements. The end of this tensor is marked with zero. It outputs in the second output port if return_inverse is false, otherwise, it outputs in the third output port. Optional.

Example

<layer ... type="Unique" ...>
<data sorted="false" return_inverse="true" return_counts="false"/>
<input>
<port id="0">
<dim>20</dim>
</port>
</input>
<output>
<port id="0">
<dim>20</dim>
</port>
<port id="1">
<dim>20</dim>
</port>
</output>
</layer>

Unsqueeze Layer

Name: Unsqueeze

Category: Layer

Short description: Unsqueeze adds dimensions of size 1 to the first input tensor. The second input value specifies a list of dimensions that will be inserted. Indices specify dimensions in the output tensor.

Parameters: Unsqueeze layer doesn't have parameters.

Inputs:

  • 1: Multidimensional input blob. Required.
  • 2: OD or 1D tensor with dimensions indices to be set to 1. Values could be negative. Indices could be integer or float values.

Example

Example 1:

<layer type="Unsqueeze">
<input>
<port id="0">
<dim>2</dim>
<dim>3</dim>
</port>
</input>
<input>
<port id="1">
<dim>2</dim> <!-- value is [0, 3] -->
</port>
</input>
<output>
<port id="2">
<dim>1</dim>
<dim>2</dim>
<dim>3</dim>
<dim>1</dim>
</port>
</output>
</layer>

Example 2: (unsqueeze 0D tensor (constant) to 1D tensor)

<layer type="Unsqueeze">
<input>
<port id="0">
</port>
</input>
<input>
<port id="1">
<dim>1</dim> <!-- value is [0] -->
</port>
</input>
<output>
<port id="2">
<dim>1</dim>
</port>
</output>
</layer>

Unique Layer

Back to top

Name: Unique

Category: Layer

Short description: Unique finds unique elements in 1-D tensor.

Detailed description: Reference

Parameters: Unique layer parameters should be specified in the data node, which is a child of the layer node.

  • Parameter name: sorted
    • Description: If sorted is equal to true, the unique elements in the output are sorted in ascending order. Otherwise, all of the unique elements are sorted in the same order as they occur in the input.
    • Range of values: true or false
    • Type: string
    • Required: yes
  • Parameter name: return_inverse
    • Description: If return_inverse is equal to true, the layer outputs the indices. Otherwise, it does not.
    • Range of values: true or false
    • Type: string
    • Required: yes
  • Parameter name: return_counts
    • Description: If return_counts is equal to true, the layer outputs the counts for each unique element. Otherwise, it does not.
    • Range of values: true or false
    • Type: string
    • Required: yes

Input:

  • 1: 1-D tensor. Input tensor. Required.

Outputs:

  • 1: 1-D tensor. Tensor of all unique elements from the input tensor. As a number of unique elements can be less than a size of the input, the end of this tensor is marked with the latest unique element. Required.
  • 2: 1-D tensor. Tensor of indices of unique elements of the first output that can be used to reconstruct the input. The size of this tensor is equal to the input size. It outputs in the second output port. Optional.
  • 3: 1-D tensor. Tensor of counts of occurrences of each unique element in the input. It has the same size as the output with unique elements. The end of this tensor is marked with zero. It outputs in the second output port if return_inverse is false, otherwise, it outputs in the third output port. Optional.

Example

<layer ... type="Unique" ...>
<data sorted="false" return_inverse="true" return_counts="false"/>
<input>
<port id="0">
<dim>20</dim>
</port>
</input>
<output>
<port id="0">
<dim>20</dim>
</port>
<port id="1">
<dim>20</dim>
</port>
</output>
</layer>