MorphPool: Efficient Non-linear Pooling & Unpooling in CNNs

TL;DR

MorphPool introduces a generalized morphological pooling and unpooling method for CNNs, enhancing performance and reducing parameters in encoder-decoder architectures for pixel-level tasks.

Contribution

This paper proposes MorphPool, a novel morphological pooling and unpooling framework that extends traditional max pooling, improving accuracy and efficiency in neural networks.

Findings

Improved predictive performance on multiple datasets

Reduced parameter counts compared to traditional methods

Effective in encoder-decoder architectures for pixel-level tasks

Abstract

Pooling is essentially an operation from the field of Mathematical Morphology, with max pooling as a limited special case. The more general setting of MorphPooling greatly extends the tool set for building neural networks. In addition to pooling operations, encoder-decoder networks used for pixel-level predictions also require unpooling. It is common to combine unpooling with convolution or deconvolution for up-sampling. However, using its morphological properties, unpooling can be generalised and improved. Extensive experimentation on two tasks and three large-scale datasets shows that morphological pooling and unpooling lead to improved predictive performance at much reduced parameter counts.