【Pooling】What is GeM(Generalized Mean Pooling)


1. GeM

GeM(Generalized Mean Pooling) is like middle method between average pooling and max pooling.

The formula is this:
\left( \dfrac{1}{n} \sum_{i=1}^n x_i^p \right)^{\frac{1}{p}}

Features of GeM

The parameter p can be learned during the training process, which allows the model to adapt the pooling mechanism to better suit the dataset and task at hand. This adaptability can lead to improvements in model performance because the pooling layer can learn to focus on the most informative features.

By the way, GeM will be max pooling when p \rightarrow \infty, and will be average pooling when p = 1

2. Conclution

Gem is learnable pooling. It provide otimal pooling for model.