Learning ordered pooling weights in image classification

TL;DR

This paper introduces a learnable ordered weighted average pooling method that enhances feature aggregation in image classification, outperforming traditional max and average pooling in CNNs and Bag-of-Words models.

Contribution

It proposes a novel approach to learn pooling weights using OWA operators, improving discriminative power in image classification tasks.

Findings

OWA pooling outperforms max and average pooling in experiments.

The method is effective in both Bag-of-Words and CNN frameworks.

Learned weights adapt to different image features.

Abstract

Spatial pooling is an important step in computer vision systems like Convolutional Neural Networks or the Bag-of-Words method. The spatial pooling purpose is to combine neighbouring descriptors to obtain a single descriptor for a given region (local or global). The resultant combined vector must be as discriminant as possible, in other words, must contain relevant information, while removing irrelevant and confusing details. Maximum and average are the most common aggregation functions used in the pooling step. To improve the aggregation of relevant information without degrading their discriminative power for image classification, we introduce a simple but effective scheme based on Ordered Weighted Average (OWA) aggregation operators. We present a method to learn the weights of the OWA aggregation operator in a Bag-of-Words framework and in Convolutional Neural Networks, and provide an extensive evaluation showing that OWA based pooling outperforms classical aggregation operators.