Spatially-Adaptive Filter Units for Deep Neural Networks

TL;DR

This paper introduces Displaced Aggregation Units (DAUs), a learnable, spatially-adaptive filter unit for deep neural networks that improves efficiency and flexibility over traditional fixed-grid filters, demonstrated on classification and segmentation tasks.

Contribution

The paper proposes DAUs, a novel learnable filter unit that adapts spatially within deep networks, reducing parameters and improving convergence without hand-crafted design.

Findings

DAUs achieve comparable accuracy to traditional filters.

DAUs enable faster convergence in training.

DAUs reduce parameter count by up to 3 times.

Abstract

Classical deep convolutional networks increase receptive field size by either gradual resolution reduction or application of hand-crafted dilated convolutions to prevent increase in the number of parameters. In this paper we propose a novel displaced aggregation unit (DAU) that does not require hand-crafting. In contrast to classical filters with units (pixels) placed on a fixed regular grid, the displacement of the DAUs are learned, which enables filters to spatially-adapt their receptive field to a given problem. We extensively demonstrate the strength of DAUs on a classification and semantic segmentation tasks. Compared to ConvNets with regular filter, ConvNets with DAUs achieve comparable performance at faster convergence and up to 3-times reduction in parameters. Furthermore, DAUs allow us to study deep networks from novel perspectives. We study spatial distributions of DAU filters and analyze the number of parameters allocated for spatial coverage in a filter.