Style-Coherent Multi-Modality Image Fusion
Keywords: Multi-modality, Image Fusion, Style-based Learning, Self-supervised Learning
Abstract: Multi-modality image fusion (MMIF) integrates heterogeneous images from diverse sensors. However, existing MMIF methods often overlook significant style discrepancies, such as saturation and resolution differences between modalities, resulting in overly smooth features in certain modalities. This tendency causes models to misjudge and disregard potentially crucial content. To address this issue, this paper proposes a novel style-coherent multi-modality fusion model that adeptly merges heterogeneous styled features from various modalities. Specifically, the proposed style-normalized fusion module progressively supplements the complete content structure by merging style-normalized features during cross-modal feature extraction. Meanwhile, a style-alignment fusion module is developed to align different feature representations across modalities, ensuring consistency. Additionally, to better preserve information and emphasize critical patterns during fusion, an adaptive reconstruction loss is applied to multi-modal images transformed into a unified image domain, enforcing mapping to a consistent modality representation. Extensive experiments validate that our method outperforms existing approaches on multiple MMIF tasks and exhibits greater potential to facilitate downstream applications.
Supplementary Material: pdf
Primary Area: applications to computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 2556
Loading