Pooling Operations in Deep Learning: From “Invariable” to “Variable”

Table 3

Summary of variable of pooling domain.

Pooling method	Characteristic	Sketch map
Region of interest pooling [17]	ROI pooling operation converts feature maps within a region of interest of any size into feature maps of fixed size
Fractional max pooling [19]	Fractional max pooling operation allows the pooling domain to be noninteger values and reduces overfitting
Strip pooling [21]	Strip pooling operation is easy to establish remote dependencies between discrete distributed regions. It can capture local details and can be easily embedded into any building block
Chunk-max pooling [22]	Chunk-max pooling operation retains the relative order information of multiple local max eigenvalues but does not retain the absolute position information
Multiscale orderless pooling [23]	Multiscale orderless pooling operation improves the invariance of neural network activations without reducing its resolution