Spatially-Adaptive Filter Units for Deep Neural Networks

Domen Tabernik, Matej Kristan, Aleš Leonardis
11/30/2017
cs.CV

Official code: skokec/DAU-ConvNet (21 stars, last commit 2026-04-20).

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.

View on arXivView PDF

Code Implementations(1)

skokec/DAU-ConvNetOfficial100%

Displaced Aggregation Units for Convolutional Networks from "Spatially-Adaptive Filter Units for Deep Neural Networks" paper

215C++, CudaMar 26, 20186 months ago

Similar Papers

Spatially-Adaptive Filter Units for Compact and Efficient Deep Neural Networks

92% match1 repo212019

Multi-Scale Context Aggregation by Dilated Convolutions

83% match1 repo7842015

DiCENet: Dimension-wise Convolutions for Efficient Networks

83% match1 repo4132019

DeformUX-Net: Exploring a 3D Foundation Backbone for Medical Image Segmentation with Depthwise Deformable Convolution

83% match1 repo222023

FCN+: Global Receptive Convolution Makes FCN Great Again

83% match1 repo12023

Detail-Preserving Pooling in Deep Networks

83% match1 repo1182018

Cite this paper

@article{tabernik2017spatiallyadaptive,
  title  = {Spatially-Adaptive Filter Units for Deep Neural Networks},
  author = {Domen Tabernik and Matej Kristan and Aleš Leonardis},
  year   = {2017},
  eprint = {1711.11473},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1711.11473}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/1711.11473.svg)](https://codeofpaper.com/paper/1711.11473)

Discussion