Enhancing Feature Representation of Infrared Foundation Model Using Off‐the‐Shelf Vision Foundation Models
Electronics Letters · 2026-01 · 62 (1)
DOI: 10.1049/ell2.70658 ↗Kurz gefasst
Fachartikel in der Zeitschrift Electronics Letters, online erschienen am 06.07.2026. Verfasst von Jun‐Hyung Kim und 1 weiteren Autor:innen. Die Arbeit nennt 33 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0). Gefördert von: National Research Foundation of Korea.
Abstract (Originalsprache)
This paper proposes two pre‐training methods that leverage off‐the‐shelf vision foundation models to improve the feature representations of an infrared foundation model. Unlike conventional vision foundation models trained on natural images, infrared foundation models are typically pre‐trained on far fewer images, which can become a major bottleneck to learning robust and generalizable representations. To address this limitation, the proposed methods build upon the InfMAE framework and transfer generic knowledge from large‐scale models such as DINOv2 and CLIP to the infrared foundation model during pre‐training. The first method employs classification‐token guidance for global representation transfer, while the second utilizes patch‐token guidance to align local features through a dual‐task design. Both approaches address the modality gap between visible and infrared imagery by selectively transferring relevant information during masked image modeling. Experiments demonstrate that models pre‐trained by the proposed methods show improved performance on two downstream tasks and 4 datasets.
Förderung
- National Research Foundation of Korea
Zitieren
APA
Kim, J., & Kwon, G. (2026). Enhancing Feature Representation of Infrared Foundation Model Using Off‐the‐Shelf Vision Foundation Models. Electronics Letters, 62(1). https://doi.org/10.1049/ell2.70658
Vancouver
Kim J, Kwon G. Enhancing Feature Representation of Infrared Foundation Model Using Off‐the‐Shelf Vision Foundation Models. Electronics Letters. 2026;62(1). doi:10.1049/ell2.70658
BibTeX
@article{Kim2026,
author = {Kim, Jun‐Hyung and Kwon, Goo‐Rak},
title = {Enhancing Feature Representation of Infrared Foundation Model Using Off‐the‐Shelf Vision Foundation Models},
journal = {Electronics Letters},
year = {2026},
volume = {62},
number = {1},
doi = {10.1049/ell2.70658}
}
EUVIMED