LCIP: Loss-controlled inverse projection of high-dimensional image data

Yu Wang; Frederik L. Dennig; Michael Behrisch; Alexandru Telea

doi:10.1177/14738716261455130

LCIP: Loss-controlled inverse projection of high-dimensional image data

Yu Wang^*
, Frederik L. Dennig
, Michael Behrisch
, Alexandru Telea^*

^*Corresponding author for this work

Sub Visualisation and Graphics

University of Konstanz

Research output: Contribution to journal › Article › Academic › peer-review

Abstract

Projections, also known as dimensionality reduction methods, (Formula presented) aim to map high-dimensional data to 2D scatterplots for visual exploration. Inverse projection methods (Formula presented) aim to map this 2D space to the data space to support tasks such as data augmentation, classifier analysis, and data imputation. Current (Formula presented) methods suffer from a fundamental limitation – they can only generate a fixed surface-like structure in data space, which poorly covers the richness of this space. We address this by a new method that ‘sweeps’ the data space by a surface that is not fixed but under user control. Our method works generically for any technique (Formula presented) and dataset, is controlled by two intuitive user-set parameters, and is simple to implement. We demonstrate it by an extensive application involving image manipulation for style transfer.

Original language	English
Journal	Information Visualization
DOIs	https://doi.org/10.1177/14738716261455130
Publication status	E-pub ahead of print - 23 Jun 2026

Bibliographical note

Publisher Copyright:
© The Author(s) 2026. This article is distributed under the terms of the Creative Commons Attribution 4.0 License (https://creativecommons.org/licenses/by/4.0/) which permits any use, reproduction and distribution of the work without further permission provided the original work is attributed as specified on the SAGE and Open Access page (https://us.sagepub.com/en-us/nam/open-access-at-sage).

Keywords

adversarial training
disentanglement
inverse projection
multidimensional data visualization

Access to Document

10.1177/14738716261455130Licence: CC BY

Cite this

@article{f5c0312d85034de2a79eb6e4448afa05,

title = "LCIP: Loss-controlled inverse projection of high-dimensional image data",

abstract = "Projections, also known as dimensionality reduction methods, (Formula presented) aim to map high-dimensional data to 2D scatterplots for visual exploration. Inverse projection methods (Formula presented) aim to map this 2D space to the data space to support tasks such as data augmentation, classifier analysis, and data imputation. Current (Formula presented) methods suffer from a fundamental limitation – they can only generate a fixed surface-like structure in data space, which poorly covers the richness of this space. We address this by a new method that {\textquoteleft}sweeps{\textquoteright} the data space by a surface that is not fixed but under user control. Our method works generically for any technique (Formula presented) and dataset, is controlled by two intuitive user-set parameters, and is simple to implement. We demonstrate it by an extensive application involving image manipulation for style transfer.",

keywords = "adversarial training, disentanglement, inverse projection, multidimensional data visualization",

author = "Yu Wang and Dennig, \{Frederik L.\} and Michael Behrisch and Alexandru Telea",

note = "Publisher Copyright: {\textcopyright} The Author(s) 2026. This article is distributed under the terms of the Creative Commons Attribution 4.0 License (https://creativecommons.org/licenses/by/4.0/) which permits any use, reproduction and distribution of the work without further permission provided the original work is attributed as specified on the SAGE and Open Access page (https://us.sagepub.com/en-us/nam/open-access-at-sage).",

year = "2026",

month = jun,

day = "23",

doi = "10.1177/14738716261455130",

language = "English",

journal = "Information Visualization",

issn = "1473-8716",

publisher = "SAGE Publications Inc.",

}

TY - JOUR

T1 - LCIP

T2 - Loss-controlled inverse projection of high-dimensional image data

AU - Wang, Yu

AU - Dennig, Frederik L.

AU - Behrisch, Michael

AU - Telea, Alexandru

N1 - Publisher Copyright: © The Author(s) 2026. This article is distributed under the terms of the Creative Commons Attribution 4.0 License (https://creativecommons.org/licenses/by/4.0/) which permits any use, reproduction and distribution of the work without further permission provided the original work is attributed as specified on the SAGE and Open Access page (https://us.sagepub.com/en-us/nam/open-access-at-sage).

PY - 2026/6/23

Y1 - 2026/6/23

N2 - Projections, also known as dimensionality reduction methods, (Formula presented) aim to map high-dimensional data to 2D scatterplots for visual exploration. Inverse projection methods (Formula presented) aim to map this 2D space to the data space to support tasks such as data augmentation, classifier analysis, and data imputation. Current (Formula presented) methods suffer from a fundamental limitation – they can only generate a fixed surface-like structure in data space, which poorly covers the richness of this space. We address this by a new method that ‘sweeps’ the data space by a surface that is not fixed but under user control. Our method works generically for any technique (Formula presented) and dataset, is controlled by two intuitive user-set parameters, and is simple to implement. We demonstrate it by an extensive application involving image manipulation for style transfer.

AB - Projections, also known as dimensionality reduction methods, (Formula presented) aim to map high-dimensional data to 2D scatterplots for visual exploration. Inverse projection methods (Formula presented) aim to map this 2D space to the data space to support tasks such as data augmentation, classifier analysis, and data imputation. Current (Formula presented) methods suffer from a fundamental limitation – they can only generate a fixed surface-like structure in data space, which poorly covers the richness of this space. We address this by a new method that ‘sweeps’ the data space by a surface that is not fixed but under user control. Our method works generically for any technique (Formula presented) and dataset, is controlled by two intuitive user-set parameters, and is simple to implement. We demonstrate it by an extensive application involving image manipulation for style transfer.

KW - adversarial training

KW - disentanglement

KW - inverse projection

KW - multidimensional data visualization

UR - https://www.scopus.com/pages/publications/105042582566

U2 - 10.1177/14738716261455130

DO - 10.1177/14738716261455130

M3 - Article

AN - SCOPUS:105042582566

SN - 1473-8716

JO - Information Visualization

JF - Information Visualization

ER -

LCIP: Loss-controlled inverse projection of high-dimensional image data

Abstract

Bibliographical note

Keywords

Access to Document

Other files and links

Fingerprint

Cite this