Spatially-Adaptive Filter Units for Compact and Efficient Deep Neural Networks (2019-02-20T00:00:00.000000Z)

TL;DR

DAUs provide a seamless substitution of convolutional filters in existing state-of-the-art architectures, which are demonstrated on AlexNet, ResNet50, ResNet101, DeepLab and SRN-DeblurNet and SRN-DeblurNet.

Abstract

Convolutional neural networks excel in a number of computer vision tasks. One of their most crucial architectural elements is the effective receptive field size, which has to be manually set to accommodate a specific task. Standard solutions involve large kernels, down/up-sampling and dilated convolutions. These require testing a variety of dilation and down/up-sampling factors and result in non-compact networks and large number of parameters. We address this issue by proposing a new convolution filter composed of displaced aggregation units (DAU). DAUs learn spatial displacements and adapt the receptive field sizes of individual convolution filters to a given problem, thus reducing the need for hand-crafted modifications. DAUs provide a seamless substitution of convolutional filters in existing state-of-the-art architectures, which we demonstrate on AlexNet, ResNet50, ResNet101, DeepLab and SRN-DeblurNet. The benefits of this design are demonstrated on a variety of computer vision tasks and datasets, such as image classification (ILSVRC 2012), semantic segmentation (PASCAL VOC 2011, Cityscape) and blind image de-blurring (GOPRO). Results show that DAUs efficiently allocate parameters resulting in up to 4×\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times $$\end{document} more compact networks in terms of the number of parameters at similar or better performance.

Authors

Domen Tabernik

5 papers

A. Leonardis

9 papers

M. Kristan

9 papers

TL;DR

Abstract

Authors

References50 items

Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation

Scale-Recurrent Network for Deep Image Deblurring

Spatially-Adaptive Filter Units for Deep Neural Networks

Deep Dual Learning for Semantic Image Segmentation

Rethinking Atrous Convolution for Semantic Image Segmentation

Dilated Residual Networks

Gabor Convolutional Networks

Fully Convolutional Networks for Semantic Segmentation

Active Convolution: Learning the Shape of Convolution for Image Classification

Mask R-CNN

Deformable Convolutional Networks

YOLO9000: Better, Faster, Stronger

Deep Multi-scale Convolutional Neural Network for Dynamic Scene Deblurring

Understanding the Effective Receptive Field in Deep Convolutional Neural Networks

Aggregated Residual Transformations for Deep Neural Networks

Towards deep compositional networks

DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs

Structured Receptive Fields in CNNs

The Cityscapes Dataset for Semantic Urban Scene Understanding

Identity Mappings in Deep Residual Networks

SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <1MB model size

Deep Residual Learning for Image Recognition

SOPHIE velocimetry of Kepler transit candidates XVII. The physical properties of giant exoplanets within 400 days of period

Fast-classifying, high-accuracy spiking deep networks through weight and threshold balancing

U-Net: Convolutional Networks for Biomedical Image Segmentation

Adam: A Method for Stochastic Optimization

Fully convolutional networks for semantic segmentation

Very Deep Convolutional Networks for Large-Scale Image Recognition

ImageNet Large Scale Visual Recognition Challenge

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

BING: Binarized normed gradients for objectness estimation at 300fps

Speeding up Convolutional Neural Networks with Low Rank Expansions

Understanding Deep Architectures using a Recursive Convolutional Network

Mastering the Discrete Fourier Transform in One, Two or Several Dimensions

Recurrent Processing during Object Recognition

ImageNet classification with deep convolutional neural networks

Invariant Scattering Convolution Networks

Semantic contours from inverse detectors

Understanding the difficulty of training deep feedforward neural networks

Structure-Aware Convolutional Neural Networks

The Cityscapes Dataset

Medical Image Computing and Computer-Assisted Intervention -- MICCAI 2015

Learning Multiple Layers of Features from Tiny Images

The PASCAL Visual Object Classes Challenge

The Pascal Visual Object Classes Challenge 2006 ( VOC 2006 ) Results

Pascal Visual Object Classes Challenge Results

GradientBased Learning Applied to Document Recognition

Gradient-based learning applied to document recognition

A real-time algorithm for signal analysis with the help of the wavelet transform

Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations

Field of Study

Journal Information

Name

Page

Volume

Venue Information

Name

Type

URL

Alternate Names