Pooling Kernels Group¶

The Pooling Group describes operations which divide input features into sub-frames and applies a function with scalar output on each of them. Generally, this results in a feature map with reduced, withheld, or emphasized key features.

Functions in this group use the mli_pool_cfg structure, defined as:

typedef struct {
   uint8_t kernel_width;
   uint8_t kernel_height;
   uint8_t stride_width;
   uint8_t stride_height;
   uint8_t padding_left;
   uint8_t padding_right;
   uint8_t padding_top;
   uint8_t padding_bottom;
} mli_pool_cfg;

mli_pool_cfg structure field description¶
Field name	Type	Description
`kernel_width`	`uint8_t`	Width of the pooling kernel.
`kernel_height`	`uint8_t`	Height of the pooling kernel.
`stride_width`	`uint8_t`	Stride of filter across width dimension of input; is the step in the input tensor in the width dimension to the next filter.
`stride_height`	`uint8_t`	Stride of filter across height dimension of input; is the step in the input tensor in the height dimension to the next filter.
`padding_left`	`uint8_t`	Number of zero points implicitly added to the left of input (width dimension)
`padding_right`	`uint8_t`	Number of zero points implicitly added to the right of input (width dimension)
`padding_top`	`uint8_t`	Number of zero points implicitly added to the top of input (height dimension)
`padding_bottom`	`uint8_t`	Number of zero points implicitly added to the bottom of input (height dimension)

For all kernels in this group, spatial dimensions of in and out tensors (Width and Height) must comply with the following system of equations:

(1)¶\[ \begin{align}\begin{aligned}\begin{cases}\\\hat{Wi} = {Wi}+padding\_left+padding\_right\\\hat{Hi} = {Hi}+padding\_top+padding\_bottom\\{Wo}*{stride\_width} = \hat{Wi}-{kernel\_width}+{stride\_width}\\{Ho}*{stride\_height} = \hat{Hi}-{kernel\_height}+{stride\_height}\\\end{cases}\end{aligned}\end{align} \]

Where:

\(\hat{Wi}\), \(\hat{Hi}\) - effective in feature map width and height after applying \(padding\_*\) to the original width (\(Wi\)) and height (\(Hi\)).

\(Wo\), \(Ho\) - out feature map width and height.