Gaussian Primitive Fusion merges multimodal images without resampling
An arXiv paper proposes fitting infrared and visible images into a shared continuous Gaussian space, avoiding the resampling step that can blur fused results.
In plain words
Cameras that capture different kinds of light, such as infrared and visible, often record images at different resolutions. To combine them, most methods first stretch the images to a common size, which can add blur or blocky errors. The proposed method instead fits both into one continuous model and renders the combined image at the sharpest available resolution.
What was announced
Multimodal image fusion combines observations from different modalities into a single representation, but most existing methods estimate each modality's contribution on a shared discrete grid. When sources have different native resolutions, building that grid usually requires resampling, which the paper's authors say can introduce artifacts and influence later fusion decisions. Gaussian Primitive Fusion (GPF) instead reformulates heterogeneous-resolution fusion in a shared continuous Gaussian space, fitting native-resolution observations with shared Gaussian geometry and modality-specific content representations.
Once optimized, the fused representation is rendered at the finest native resolution without resampling the source images. The authors report that, with heterogeneous-resolution inputs, GPF produces clearer target contours and finer textures with less blurring and blocking than resample-then-fuse pipelines, and achieves the best results on multiple infrared-visible benchmarks against state-of-the-art methods using interpolation or super-resolution frontends. These claims come from the paper's abstract, which reports the authors' own evaluations rather than independent testing.
What it does not show
- The evidence is the paper's abstract only; the full method details, experiments and peer review status are not covered.
Sources
Source-support review completed 20 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: arXiv: abstract.
Updates and corrections
No corrections so far.
First published 20 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.