Copyright © 2025 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{182043,
        author = {Rahul Vishwas Pimpale},
        title = {CrossFuse: Robust IR–Visible Fusion via Self Supervision with Top-k Alignment},
        journal = {International Journal of Innovative Research in Technology},
        year = {2025},
        volume = {12},
        number = {2},
        pages = {664-671},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=182043},
        abstract = {In multimodal image fusion, robust generalization across diverse environments remains a significant challenge—especially under label-scarce conditions and out-of-distribution (OOD) shifts. We propose CrossFuse, a novel self-supervised learning (SSL) framework for infrared (IR) and visible image fusion, combining multi-view augmentations with a Top-k Selective Vision Alignment (SVA) mechanism. CrossFuse leverages weakly aggressive augmentations to maintain modality integrity while encouraging robust feature interactions. At its core, CrossFuse introduces a cross-modal contrastive loss with Top-k mining, enabling adaptive feature selection and improved cross-sensor alignment. Through extensive experiments on challenging benchmarks such as FLIR ADAS and MFNet, CrossFuse consistently outperforms existing fusion techniques in both in-distribution and OOD scenarios. Our approach is fully label-free, enabling scalable and generalizable multimodal training. This work paves the way toward more resilient sensor fusion systems, with potential implications in autonomous navigation, remote sensing, and surveillance.},
        keywords = {Multimodal Image Fusion, Self-Supervised Learning (SSL), Top-k Vision Alignment, Cross-Modal Contrastive Learning.},
        month = {July},
        }