Viet-Anh on Software Logo

What is: Generalized Mean Pooling?

Year2000
Data SourceCC BY-SA - https://paperswithcode.com

Generalized Mean Pooling (GeM) computes the generalized mean of each channel in a tensor. Formally:

e=[(1Ω_uΩxp_cu)1p]_c=1,,C\textbf{e} = \left[\left(\frac{1}{|\Omega|}\sum\_{u\in{\Omega}}x^{p}\_{cu}\right)^{\frac{1}{p}}\right]\_{c=1,\cdots,C}

where p>0p > 0 is a parameter. Setting this exponent as p>1p > 1 increases the contrast of the pooled feature map and focuses on the salient features of the image. GeM is a generalization of the average pooling commonly used in classification networks (p=1p = 1) and of spatial max-pooling layer (p=p = \infty).

Source: MultiGrain

Image Source: Eva Mohedano