Any2Any: Unified Arbitrary Modality Translation for Remote Sensing

Multi-modal remote sensing imagery provides complementary observations of the same geographic scene, yet such observations are frequently incomplete in practice. Existing cross-modal translation methods treat each modality pair as an independent task, resulting in quadratic complexity and limited generalization to unseen modality combinations. We formulate Any-to-Any translation as inference over a shared latent representation of the scene, where different modalities correspond to partial observations of the same underlying semantics. Based on this formulation, we propose Any2Any, a unified latent diffusion framework that projects heterogeneous inputs into a geometrically aligned latent space. Such structure performs anchored latent regression with a shared backbone, decoupling modality-specific representation learning from semantic mapping. Moreover, lightweight target-specific residual adapters are used to correct systematic latent mismatches without increasing inference complexity. To support learning under sparse but connected supervision, we introduce RST-1M, the first million-scale remote sensing dataset with paired observations across five sensing modalities, providing supervision anchors for any-to-any translation. Experiments across 14 translation tasks show that Any2Any consistently outperforms pairwise translation methods and exhibits strong zero-shot generalization to unseen modality pairs. Code and models are available at https://github.com/MiliLab/Any2Any.

Different types of satellite images give us unique views of the Earth, but we often don't have all the specific views we need at the same time. Existing AI methods try to fill in these gaps by translating one specific type of image into another, but they treat every pair of image types as a completely separate problem, making them highly inefficient and hard to scale. To solve this, we created Any2Any, a unified AI system that can translate any type of satellite image into any other type. Instead of learning specific pairs, Any2Any works by understanding the core, underlying features of a geographic scene, recognizing that different images are simply different ways of looking at the exact same reality. To train this system, we also built RST-1M, a massive new dataset containing a million examples of paired satellite views across five different sensing modalities. Our experiments demonstrate that Any2Any consistently outperforms existing methods and even exhibits strong generalization to image pairs it has never seen during training.