Alias-Free Convnets: Fractional Shift Invariance via Polynomial Activations

Hagay Michaeli; Tomer Michaeli; Daniel Soudry

doi:10.1109/CVPR52729.2023.01567

Alias-Free Convnets: Fractional Shift Invariance via Polynomial Activations

Hagay Michaeli, Tomer Michaeli, Daniel Soudry

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

1 Scopus citations

Abstract

Although CNNs are believed to be invariant to translations, recent works have shown this is not the case due to aliasing effects that stem from down-sampling layers. The existing architectural solutions to prevent the aliasing effects are partial since they do not solve those effects that originate in non-linearities. We propose an extended antialiasing method that tackles both down-sampling and nonlinear layers, thus creating truly alias-free, shift-invariant CNNs11Our code is available at github.com/hmichaeli/alias-free-convnets/.. We show that the presented model is invariant to integer as well as fractional (i.e., sub-pixel) translations, thus outperforming other shift-invariant methods in terms of robustness to adversarial translations.

Original language	English
Title of host publication	Proceedings - 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2023
Pages	16333-16342
Number of pages	10
ISBN (Electronic)	9798350301298
DOIs	https://doi.org/10.1109/CVPR52729.2023.01567
State	Published - 2023
Event	2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2023 - Vancouver, Canada Duration: 18 Jun 2023 → 22 Jun 2023

Publication series

Name	Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition
Volume	2023-June
ISSN (Print)	1063-6919

Conference

Conference	2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2023
Country/Territory	Canada
City	Vancouver
Period	18/06/23 → 22/06/23

Keywords

Deep learning architectures and techniques

ASJC Scopus subject areas

Software
Computer Vision and Pattern Recognition

Access to Document

10.1109/CVPR52729.2023.01567

Cite this

Michaeli, H., Michaeli, T., & Soudry, D. (2023). Alias-Free Convnets: Fractional Shift Invariance via Polynomial Activations. In Proceedings - 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2023 (pp. 16333-16342). (Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition; Vol. 2023-June). https://doi.org/10.1109/CVPR52729.2023.01567

@inproceedings{9bafb5a117ce4ba68da2cc83d6b588fe,

title = "Alias-Free Convnets: Fractional Shift Invariance via Polynomial Activations",

abstract = "Although CNNs are believed to be invariant to translations, recent works have shown this is not the case due to aliasing effects that stem from down-sampling layers. The existing architectural solutions to prevent the aliasing effects are partial since they do not solve those effects that originate in non-linearities. We propose an extended antialiasing method that tackles both down-sampling and nonlinear layers, thus creating truly alias-free, shift-invariant CNNs11Our code is available at github.com/hmichaeli/alias-free-convnets/.. We show that the presented model is invariant to integer as well as fractional (i.e., sub-pixel) translations, thus outperforming other shift-invariant methods in terms of robustness to adversarial translations.",

keywords = "Deep learning architectures and techniques",

author = "Hagay Michaeli and Tomer Michaeli and Daniel Soudry",

note = "Publisher Copyright: {\textcopyright} 2023 IEEE.; 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2023 ; Conference date: 18-06-2023 Through 22-06-2023",

year = "2023",

doi = "10.1109/CVPR52729.2023.01567",

language = "אנגלית",

series = "Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition",

pages = "16333--16342",

booktitle = "Proceedings - 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2023",

}

Michaeli, H, Michaeli, T & Soudry, D 2023, Alias-Free Convnets: Fractional Shift Invariance via Polynomial Activations. in Proceedings - 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2023. Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, vol. 2023-June, pp. 16333-16342, 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2023, Vancouver, Canada, 18/06/23. https://doi.org/10.1109/CVPR52729.2023.01567

Alias-Free Convnets: Fractional Shift Invariance via Polynomial Activations. / Michaeli, Hagay; Michaeli, Tomer ; Soudry, Daniel.
Proceedings - 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2023. 2023. p. 16333-16342 (Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition; Vol. 2023-June).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Alias-Free Convnets

T2 - 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2023

AU - Michaeli, Hagay

AU - Michaeli, Tomer

AU - Soudry, Daniel

PY - 2023

Y1 - 2023

N2 - Although CNNs are believed to be invariant to translations, recent works have shown this is not the case due to aliasing effects that stem from down-sampling layers. The existing architectural solutions to prevent the aliasing effects are partial since they do not solve those effects that originate in non-linearities. We propose an extended antialiasing method that tackles both down-sampling and nonlinear layers, thus creating truly alias-free, shift-invariant CNNs11Our code is available at github.com/hmichaeli/alias-free-convnets/.. We show that the presented model is invariant to integer as well as fractional (i.e., sub-pixel) translations, thus outperforming other shift-invariant methods in terms of robustness to adversarial translations.

AB - Although CNNs are believed to be invariant to translations, recent works have shown this is not the case due to aliasing effects that stem from down-sampling layers. The existing architectural solutions to prevent the aliasing effects are partial since they do not solve those effects that originate in non-linearities. We propose an extended antialiasing method that tackles both down-sampling and nonlinear layers, thus creating truly alias-free, shift-invariant CNNs11Our code is available at github.com/hmichaeli/alias-free-convnets/.. We show that the presented model is invariant to integer as well as fractional (i.e., sub-pixel) translations, thus outperforming other shift-invariant methods in terms of robustness to adversarial translations.

KW - Deep learning architectures and techniques

UR - http://www.scopus.com/inward/record.url?scp=85167738577&partnerID=8YFLogxK

U2 - 10.1109/CVPR52729.2023.01567

DO - 10.1109/CVPR52729.2023.01567

M3 - ???researchoutput.researchoutputtypes.contributiontobookanthology.conference???

AN - SCOPUS:85167738577

T3 - Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition

SP - 16333

EP - 16342

BT - Proceedings - 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2023

Y2 - 18 June 2023 through 22 June 2023

ER -

Alias-Free Convnets: Fractional Shift Invariance via Polynomial Activations

Abstract

Publication series

Conference

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this