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

Table 5

Summary of multichannel pooling.

Pooling method	Characteristic	Sketch map
Intermediate value pooling [34]	It takes into account the average pooling and the max pooling, so that it has smaller model error and higher stability
Mixed pooling [35]	It solves the problem of which to choose between max pooling and average pooling, but only using one of these pooling methods still has the disadvantage of max pooling or average pooling
Spatial pyramid and global average hybrid pooling [36]	It can extract local and global information, respectively, which learns more information
Multiscale pooling	It is more flexible and can be used multiple times at the beginning, middle, or end of the network
Scalable overlapping slide pooling	It considers the saliency of features at different scales and the relationship between adjacent feature elements, so that coarse-grained, medium-grained, and fine-grained multiple features can be extracted