%20revision-blue){kind=icon}
Official repository for IterRF.
IterRF targets the challenging problem of unaligned multi-modal image fusion, where accurate spatial registration and high-quality image fusion must be optimized together rather than handled as isolated steps.
Existing approaches commonly rely on a cascaded register-then-fuse pipeline, or introduce only limited one-step feedback between registration and fusion. These designs can restrict the mutual enhancement between spatial alignment and information integration.
IterRF formulates registration and fusion within a unified optimization framework and unfolds the optimization process into alternating registration-related and fusion-related subproblems. With iterative bidirectional refinement, the model progressively improves alignment accuracy and fusion quality.
- Joint registration and fusion: optimizes spatial alignment and fused image generation in a unified framework.
- Iterative mutual enhancement: alternates between registration-oriented and fusion-oriented refinement stages.
- Optimization-inspired design: unfolds interpretable subproblems into a learnable end-to-end framework.
- Broad multi-modal applicability: designed for infrared-visible and medical multi-modal image scenarios.
- Research-friendly release plan: code, pretrained models, evaluation scripts, and visualization results will be released after paper acceptance.
The framework alternates between registration and fusion modules so that the two tasks can refine each other over multiple iterations. Registration benefits from progressively enhanced fused representations, while fusion benefits from increasingly reliable spatial correspondence.
The full project will be released upon paper acceptance. Planned materials include:
- Source code for the official IterRF implementation
- Pretrained model checkpoints
- Testing and evaluation scripts
- Training scripts and configuration files
- Visualization tools and qualitative comparison results
- Detailed usage instructions and reproducibility notes
The installation guide will be provided together with the code release.
Testing and evaluation instructions will be added after the official implementation is released.
Training scripts, configuration files, and dataset preparation instructions will be released after paper acceptance.
Pretrained checkpoints will be made available after the paper is accepted.
Quantitative results, qualitative comparisons, and visualization examples will be added in the official release.
The license will be announced before the public code release.
If you find this work useful for your research, please consider citing our paper. The BibTeX entry will be updated after publication.
For questions about IterRF, please open an issue in this repository.