SwinFusion: attention within and across the sources
Optional · paper guide 9 of 11 for DLCV 10 · Image Fusion: Complementary Images, Useful Outputs. Citation numbers refer to the reference list at the bottom of this page.
Navigate: this page and this track: Navigate: this page and this track
On this page
- Exposure fusion (Mertens, Kautz and Van Reeth)
- VIFB: a visible and infrared image fusion benchmark
- The infrared–visible fusion survey of Liu et al. (TPAMI 2025)
- DenseFuse: learn reconstruction, choose fusion in feature space
- DDFM: a diffusion prior with likelihood rectification
- FusionGAN: thermal intensity with an adversarial visible-image prior
- IFCNN: supervised fusion with a variable number of inputs
- U2Fusion: one unsupervised model across fusion tasks
- SwinFusion: attention within and across the sources
- SeAFusion: let semantic segmentation influence the fused image
- TarDAL: fuse for both appearance and object detection
Deep Learning for Computer Vision
Challenge
The useful structure in a fusion problem does not always fit inside a small convolutional neighborhood. A dim pedestrian may need context to distinguish it from a hot background; an exposure boundary may require a consistent decision across a larger object. At the same time, the two sources have different appearances. Attention within one source and attention between sources solve different problems. SwinFusion [34] builds both into a framework used for infrared–visible, visible–near-infrared, medical, multi-focus and multi-exposure fusion.
“General” here means a reusable architecture and objective design. The paper trains its models for the individual tasks; the results do not establish that one unchanged checkpoint solves all modalities. This differs from U2Fusion’s explicit continual-learning procedure.
Method

Figure source: [34], Published Figure 2 supplied by official implementation, commit 82bf7381338f90a1d056194e5b1f16cf14f01d8f. Original source.
Shallow features preserve local evidence. Each source first passes through two stride-one 3 × 3 convolutions with LeakyReLU. For a batch of aligned grayscale images of shape , this produces feature maps with shape . Here is batch size, is the feature width and are spatial dimensions. Convolutions give the attention stages a local feature representation rather than treating raw brightness as the only matching cue.
Deep extraction expands the context. Four Swin Transformer layers extract within-source features. Tokens are partitioned into windows; the default window width is eight pixels in the feature grid, with six attention heads. A window contains tokens, where is its width. For one head,
Here holds window features, are learned projections, are query, key and value matrices, is key width and is the learned relative-position bias. Softmax runs over candidate keys for each query. Multiple heads learn different comparisons; their outputs are recombined. Feed-forward layers transform token channels, while normalization and residual connections stabilize repeated updates.
Regular windows alternate with windows shifted by . The shift lets information cross a previous window boundary. This is growing context over several layers, not all-pairs global attention in one operation. Windowing makes the attention matrix depend on window area instead of the entire image area; it still costs computation and requires suitable handling of image boundaries.
Attention-guided cross-domain fusion distinguishes two interactions. The ACFM module uses intra-domain self-attention to refine each source’s own structure, followed by inter-domain attention to exchange evidence. For a direction from source 2 toward source 1, queries come from source 1 and keys/values come from source 2:
The subscript identifies which source supplies each tensor; is the imported feature update. The reverse direction is also computed. Residual paths retain each source’s information, so cross-attention is an update rather than mandatory replacement by the other modality. Two ACFM stages are used. The processed source features are concatenated and aggregated by a convolution. Matching occurs in a learned feature space, but the module is not a substitute for geometric registration.
Reconstruction turns the fused representation into pixels. Four further Swin Transformer layers process the fused deep features. Three stride-one 3 × 3 convolutions then reduce the channel count and reconstruct the output; the first two use LeakyReLU. Thus both fusion and reconstruction can use expanded context. There is no discriminator, diffusion sampler or pretrained VGG loss in this design.
Three losses encode what to retain. With fused image and aligned sources , the structural term is
SSIM is the structural-similarity index defined earlier; equal source coefficients express equal structural importance. The texture and intensity terms are
Here is the Sobel gradient operator, absolute value and maximum are elementwise, and sums absolute residuals. The task-dependent aggregation is a maximum for infrared–visible, medical and multi-focus fusion, and a mean for visible–near-infrared and multi-exposure fusion. Maximum intensity can protect bright targets; maximum gradient can preserve a strong edge from either source. Neither is a physical ground-truth image, and a maximum gradient can also favor noise.
The paper uses 128 × 128 training crops, batch size 16 and 10,000 training steps per task, with Adam and an initial learning rate of . Those are the published experimental settings, not a required recipe for arbitrary sensor data.
Results
Table I reports infrared–visible fusion FMI 0.9308, 0.6428 and SSIM 0.9676 for SwinFusion, versus U2Fusion’s 0.9135, 0.4218 and 0.9387. These are source-referenced fusion scores on that experiment. The same table gives PSNR 64.4063 for SwinFusion and 66.6532 for DenseFuse: there is no universal numerical win. An error score against source images can reward preserving one source while failing to combine complementary evidence.
The cross-task results are similarly nuanced. On CT–MRI, SwinFusion’s SSIM is 0.9497, below IFCNN’s 0.9564, although its is 0.6529 versus IFCNN’s 0.6233. The important evidence is the trade-off between retention criteria under each modality pairing, together with the displayed outputs. These numbers do not establish clinical benefit or preservation of every diagnostically relevant structure.
Efficiency is also a trade-off. Table VI reports 0.9737 M parameters and 1.4668 s for the infrared–visible experiment, compared with IFCNN’s 0.0836 M and 0.0130 s under the paper’s timing setup. The attention model is not the fastest option simply because it uses windows. Do not transfer this latency to another resolution or device.
Ablations
Table V separates deep extraction (DE), the entire ACFM, its inter-domain interaction and each loss. Removing DE lowers infrared–visible from 0.6428 to 0.3355. Removing ACFM gives 0.4592, and removing only inter-domain interaction gives 0.5612. Under this controlled setup, learned within-source context alone does not reproduce the benefit of exchanging information between sources.
Removing the SSIM term yields infrared–visible 0.2374, removing texture loss 0.3809, and removing intensity loss 0.4742. For multi-exposure SSIM, the full model gives 0.9885; removing texture and intensity terms gives 0.8587 and 0.8231 respectively. The terms constrain different failure modes, rather than providing interchangeable penalties.
Metric disagreement remains visible inside the ablation: removing texture loss increases infrared–visible PSNR from 64.4063 to 67.0619 while reducing edge-transfer quality. Also, Table V prints the full-model infrared–visible SSIM as 0.96976, whereas Table I prints 0.9676. This guide preserves the main comparison’s value and does not silently reconcile the inconsistency.
Conclusion
SwinFusion’s contribution is the combination of within-source and between-source window attention, followed by a reconstruction stage and task-adapted source-retention losses. Its experiments support useful interaction between architecture and objective, with explicit speed and metric trade-offs. The reusable design is a framework for several fusion tasks, rather than proof of one universal model or one universal quality score.
Problems
Window attention does not solve misregistration, source blur, missing measurements or adversarial corruption. Its losses ask the network to retain modeled source structure; they do not say which source is correct when both are degraded. Attention features can propagate a mistaken correspondence, and source-intensity objectives may retain unwanted thermal background or exposure artifacts.
The task-specific aggregation and training data are part of the method. Applying the model to hyperspectral reconstruction or a diagnostic workflow requires a different measurement and evaluation argument. For a different image-prior strategy, compare DDFM; for explicit downstream objectives, compare TarDAL.
References
- Y. Liu, X. Chen, Z. Wang, Z. J. Wang, R. K. Ward and X. Wang, “Deep learning for pixel-level image fusion: Recent advances and future prospects,” Information Fusion, vol. 42, pp. 158–173, 2018. doi
- J. Liu, G. Wu, Z. Liu, D. Wang, Z. Jiang, L. Ma, W. Zhong, X. Fan and R. Liu, “Infrared and visible image fusion: From data compatibility to task adaption,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 47, no. 4, pp. 2349–2369, 2025. doi arXiv
- J. Ma, Y. Ma and C. Li, “Infrared and visible image fusion methods and applications: A survey,” Information Fusion, vol. 45, pp. 153–178, 2019. doi
- P. E. Debevec and J. Malik, “Recovering high dynamic range radiance maps from photographs,” in Proc. SIGGRAPH, 1997, pp. 369–378. doi
- T. Mertens, J. Kautz and F. Van Reeth, “Exposure fusion: A simple and practical alternative to high dynamic range photography,” Computer Graphics Forum, vol. 28, no. 1, pp. 161–171, 2009. doi
- S. Xu, X. Wei, C. Zhang, J. Liu and J. Zhang, “MFFW: A new dataset for multi-focus image fusion,” arXiv:2002.04780, 2020. arXiv
- X. Zhang, “Multi-focus image fusion: A benchmark,” arXiv:2005.01116, 2020. arXiv
- P. J. Burt and E. H. Adelson, “The Laplacian pyramid as a compact image code,” IEEE Transactions on Communications, vol. 31, no. 4, pp. 532–540, 1983. doi
- P. J. Burt and R. J. Kolczynski, “Enhanced image capture through fusion,” in Proc. 4th International Conference on Computer Vision (ICCV), 1993, pp. 173–182. doi
- P. J. Burt and E. H. Adelson, “A multiresolution spline with application to image mosaics,” ACM Transactions on Graphics, vol. 2, no. 4, pp. 217–236, 1983. doi
- H. Li, B. S. Manjunath and S. K. Mitra, “Multisensor image fusion using the wavelet transform,” Graphical Models and Image Processing, vol. 57, no. 3, pp. 235–245, 1995. doi
- S. Li, X. Kang and J. Hu, “Image fusion with guided filtering,” IEEE Transactions on Image Processing, vol. 22, no. 7, pp. 2864–2875, 2013. doi
- K. He, J. Sun and X. Tang, “Guided image filtering,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 35, no. 6, pp. 1397–1409, 2013. doi
- B. Yang and S. Li, “Multifocus image fusion and restoration with sparse representation,” IEEE Transactions on Instrumentation and Measurement, vol. 59, no. 4, pp. 884–892, 2010. doi
- M. Nejati, S. Samavi and S. Shirani, “Multi-focus image fusion using dictionary-based sparse representation,” Information Fusion, vol. 25, pp. 72–84, 2015. doi
- Z. Wang, A. C. Bovik, H. R. Sheikh and E. P. Simoncelli, “Image quality assessment: From error visibility to structural similarity,” IEEE Transactions on Image Processing, vol. 13, no. 4, pp. 600–612, 2004. doi
- X. Zhang, P. Ye and G. Xiao, “VIFB: A visible and infrared image fusion benchmark,” in Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2020. CVF arXiv
- X. Zhang, “Benchmarking and comparing multi-exposure image fusion algorithms,” Information Fusion, vol. 74, pp. 111–131, 2021. doi arXiv
- A. M. Eskicioglu and P. S. Fisher, “Image quality measures and their performance,” IEEE Transactions on Communications, vol. 43, no. 12, pp. 2959–2965, 1995. doi
- G. Qu, D. Zhang and P. Yan, “Information measure for performance of image fusion,” Electronics Letters, vol. 38, no. 7, pp. 313–315, 2002. doi
- M. Hossny, S. Nahavandi and D. Creighton, “Comments on ‘Information measure for performance of image fusion’,” Electronics Letters, vol. 44, no. 18, pp. 1066–1067, 2008. doi
- C. S. Xydeas and V. Petrović, “Objective image fusion performance measure,” Electronics Letters, vol. 36, no. 4, pp. 308–309, 2000. doi
- G. Piella and H. Heijmans, “A new quality metric for image fusion,” in Proc. IEEE International Conference on Image Processing (ICIP), 2003, vol. 3, pp. III-173–176. doi
- Y. Han, Y. Cai, Y. Cao and X. Xu, “A new image fusion performance metric based on visual information fidelity,” Information Fusion, vol. 14, no. 2, pp. 127–135, 2013. doi
- K. Ma, K. Zeng and Z. Wang, “Perceptual quality assessment for multi-exposure image fusion,” IEEE Transactions on Image Processing, vol. 24, no. 11, pp. 3345–3356, 2015. doi
- Z. Liu, E. Blasch, Z. Xue, J. Zhao, R. Laganiere and W. Wu, “Objective assessment of multiresolution image fusion algorithms for context enhancement in night vision: A comparative study,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 34, no. 1, pp. 94–109, 2012. doi
- J. Liu, X. Fan, Z. Huang, G. Wu, R. Liu, W. Zhong and Z. Luo, “Target-aware dual adversarial learning and a multi-scenario multi-modality benchmark to fuse infrared and visible for object detection,” in Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 5802–5811. CVF arXiv M3FD
- J. Liu, Z. Liu, G. Wu, L. Ma, R. Liu, W. Zhong, Z. Luo and X. Fan, “Multi-interactive feature learning and a full-time multi-modality benchmark for image fusion and segmentation,” in Proc. IEEE/CVF International Conference on Computer Vision (ICCV), 2023, pp. 8115–8124. CVF arXiv
- Y. Zhang, Y. Liu, P. Sun, H. Yan, X. Zhao and L. Zhang, “IFCNN: A general image fusion framework based on convolutional neural network,” Information Fusion, vol. 54, pp. 99–118, 2020. doi
- H. Li and X.-J. Wu, “DenseFuse: A fusion approach to infrared and visible images,” IEEE Transactions on Image Processing, vol. 28, no. 5, pp. 2614–2623, 2019. doi arXiv
- K. R. Prabhakar, V. S. Srikar and R. V. Babu, “DeepFuse: A deep unsupervised approach for exposure fusion with extreme exposure image pairs,” in Proc. IEEE International Conference on Computer Vision (ICCV), 2017, pp. 4724–4732. doi CVF
- H. Xu, J. Ma, J. Jiang, X. Guo and H. Ling, “U2Fusion: A unified unsupervised image fusion network,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 44, no. 1, pp. 502–518, 2022. doi RoadScene
- L. Tang, J. Yuan and J. Ma, “Image fusion in the loop of high-level vision tasks: A semantic-aware real-time infrared and visible image fusion network,” Information Fusion, vol. 82, pp. 28–42, 2022. doi
- J. Ma, L. Tang, F. Fan, J. Huang, X. Mei and Y. Ma, “SwinFusion: Cross-domain long-range learning for general image fusion via Swin Transformer,” IEEE/CAA Journal of Automatica Sinica, vol. 9, no. 7, pp. 1200–1217, 2022. doi
- X. Zhang and Y. Demiris, “Visible and infrared image fusion using deep learning,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 8, pp. 10535–10554, 2023. doi
- Y. Liu, X. Chen, H. Peng and Z. Wang, “Multi-focus image fusion with a deep convolutional neural network,” Information Fusion, vol. 36, pp. 191–207, 2017. doi
- K. Ma, Z. Duanmu, H. Zhu, Y. Fang and Z. Wang, “Deep guided learning for fast multi-exposure image fusion,” IEEE Transactions on Image Processing, vol. 29, pp. 2808–2819, 2020. doi
- Z. Zhao, H. Bai, J. Zhang, Y. Zhang, S. Xu, Z. Lin, R. Timofte and L. Van Gool, “CDDFuse: Correlation-driven dual-branch feature decomposition for multi-modality image fusion,” in Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023, pp. 5906–5916. CVF arXiv
- J. Ma, W. Yu, P. Liang, C. Li and J. Jiang, “FusionGAN: A generative adversarial network for infrared and visible image fusion,” Information Fusion, vol. 48, pp. 11–26, 2019. doi · public article transcription
- H. Bai, J. Zhang, Z. Zhao, Y. Wu, L. Deng, Y. Cui, T. Feng and S. Xu, “Task-driven image fusion with learnable fusion loss,” in Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025, pp. 7457–7468. CVF arXiv
- Z. Zhao, H. Bai, Y. Zhu, J. Zhang, S. Xu, Y. Zhang, K. Zhang, D. Meng, R. Timofte and L. Van Gool, “DDFM: Denoising diffusion model for multi-modality image fusion,” in Proc. IEEE/CVF International Conference on Computer Vision (ICCV), 2023, pp. 8082–8093. CVF arXiv
- X. Yi, H. Xu, H. Zhang, L. Tang and J. Ma, “Text-IF: Leveraging semantic text guidance for degradation-aware and interactive image fusion,” in Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024, pp. 27026–27035. CVF arXiv
- D. Wang, J. Liu, X. Fan and R. Liu, “Unsupervised misaligned infrared and visible image fusion via cross-modality image generation and registration,” in Proc. International Joint Conference on Artificial Intelligence (IJCAI), 2022, pp. 3508–3515. doi arXiv
- L. Tang, Y. Deng, Y. Ma, J. Huang and J. Ma, “SuperFusion: A versatile image registration and fusion network with semantic awareness,” IEEE/CAA Journal of Automatica Sinica, vol. 9, no. 12, pp. 2121–2137, 2022. doi
- A. Toet, “The TNO multiband image data collection,” Data in Brief, vol. 15, pp. 249–251, 2017. doi
- L. Tang, J. Yuan, H. Zhang, X. Jiang and J. Ma, “PIAFusion: A progressive infrared and visible image fusion network based on illumination aware,” Information Fusion, vol. 83–84, pp. 79–92, 2022. doi MSRS
- Q. Ha, K. Watanabe, T. Karasawa, Y. Ushiku and T. Harada, “MFNet: Towards real-time semantic segmentation for autonomous vehicles with multi-spectral scenes,” in Proc. IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2017, pp. 5108–5115. doi
- X. Jia, C. Zhu, M. Li, W. Tang and W. Zhou, “LLVIP: A visible-infrared paired dataset for low-light vision,” in Proc. IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, 2021, pp. 3496–3504. CVF arXiv
- J. Cai, S. Gu and L. Zhang, “Learning a deep single image contrast enhancer from multi-exposure images,” IEEE Transactions on Image Processing, vol. 27, no. 4, pp. 2049–2062, 2018. doi
- X. Zhang, “Deep learning-based multi-focus image fusion: A survey and a comparative study,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 44, no. 9, 2022. doi
- F. Xu, J. Liu, Y. Song, H. Sun and X. Wang, “Multi-exposure image fusion techniques: A comprehensive review,” Remote Sensing, vol. 14, no. 3, art. 771, 2022. doi
- Z. Zhao, H. Bai, J. Zhang, Y. Zhang, K. Zhang, S. Xu, D. Chen, R. Timofte and L. Van Gool, “Equivariant multi-modality image fusion,” in Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024, pp. 25912–25921. CVF arXiv
- L. Tang, Y. Wang, M. Gong, Z. Li, Y. Deng, X. Yi, C. Li, H. Xu, H. Zhang and J. Ma, “VideoFusion: A spatio-temporal collaborative network for multi-modal video fusion,” in Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2026, pp. 19559–19569. CVF
- N. Yokoya, C. Grohnfeldt and J. Chanussot, “Hyperspectral and multispectral data fusion: A comparative review of the recent literature,” IEEE Geoscience and Remote Sensing Magazine, 2017. doi
- H. R. Sheikh and A. C. Bovik, “Image information and visual quality,” IEEE Transactions on Image Processing, 2006. doi
- H. Xu et al., “U2Fusion,” official implementation accompanying the TPAMI paper, 2022. code
- K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” ICLR, 2015. arXiv
- J. Kirkpatrick et al., “Overcoming catastrophic forgetting in neural networks,” Proceedings of the National Academy of Sciences, 2017. doi
- Y. Zhang et al., “IFCNN,” official implementation and errata accompanying the Information Fusion paper, 2020. code
- K. He, X. Zhang, S. Ren and J. Sun, “Deep residual learning for image recognition,” CVPR, 2016. CVF
- J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li and L. Fei-Fei, “ImageNet: A large-scale hierarchical image database,” CVPR, 2009. doi
- N. Silberman, D. Hoiem, P. Kohli and R. Fergus, “Indoor segmentation and support inference from RGBD images,” ECCV, 2012. project
- P. Dhariwal and A. Nichol, “Diffusion models beat GANs on image synthesis,” NeurIPS, 2021. arXiv
- J. Liu et al., “TarDAL,” official implementation accompanying the CVPR paper, 2022. code
- G. Jocher et al., “YOLOv5,” Ultralytics object-detection software, 2020 onward. code
- J. Ma et al., “FusionGAN,” official implementation accompanying the Information Fusion paper, 2019. code
- L. Tang, J. Yuan and J. Ma, “SeAFusion,” official implementation accompanying the Information Fusion paper, 2022. code
- T. Mertens, J. Kautz and F. Van Reeth, “Exposure Fusion,” in Proc. Pacific Graphics, 2007. paper · author project