Spatially-adaptive filter units for compact and efficient deep neural networks

Research output: Contribution to journalArticlepeer-review

Standard

Spatially-adaptive filter units for compact and efficient deep neural networks. / Tabernik, Domen; Kristan, Matej; Leonardis, Aleš.

In: International Journal of Computer Vision, Vol. 128, No. 8-9, 09.2020, p. 2049-2067.

Research output: Contribution to journalArticlepeer-review

Harvard

APA

Vancouver

Author

Bibtex

@article{f64fe4cbba89400ba1eed4e627ca9d1c,
title = "Spatially-adaptive filter units for compact and efficient deep neural networks",
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× more compact networks in terms of the number of parameters at similar or better performance.",
keywords = "Adjustable receptive fields, Compact ConvNets, Displacement units, Efficient ConvNets",
author = "Domen Tabernik and Matej Kristan and Ale{\v s} Leonardis",
year = "2020",
month = sep,
doi = "10.1007/s11263-019-01282-1",
language = "English",
volume = "128",
pages = "2049--2067",
journal = "International Journal of Computer Vision",
issn = "0920-5691",
publisher = "Springer",
number = "8-9",

}

RIS

TY - JOUR

T1 - Spatially-adaptive filter units for compact and efficient deep neural networks

AU - Tabernik, Domen

AU - Kristan, Matej

AU - Leonardis, Aleš

PY - 2020/9

Y1 - 2020/9

N2 - 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× more compact networks in terms of the number of parameters at similar or better performance.

AB - 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× more compact networks in terms of the number of parameters at similar or better performance.

KW - Adjustable receptive fields

KW - Compact ConvNets

KW - Displacement units

KW - Efficient ConvNets

UR - http://www.scopus.com/inward/record.url?scp=85077567153&partnerID=8YFLogxK

U2 - 10.1007/s11263-019-01282-1

DO - 10.1007/s11263-019-01282-1

M3 - Article

AN - SCOPUS:85077567153

VL - 128

SP - 2049

EP - 2067

JO - International Journal of Computer Vision

JF - International Journal of Computer Vision

SN - 0920-5691

IS - 8-9

ER -