🛰️ Remote Sensing

🧪 ICML2026 · 3 paper notes

📌 Same area in other venues: 📷 CVPR2026 (15) · 🔬 ICLR2026 (6) · 🤖 AAAI2026 (8) · 🧠 NeurIPS2025 (12) · 📹 ICCV2025 (11)

Any2Any: Unified Arbitrary Modality Translation for Remote Sensing

Any2Any transforms remote sensing (RS) translation between sensors like RGB, SAR, NIR, MS, and PAN from a collection of paired models into a unified latent diffusion model within a shared latent space. Leveraging the million-scale RST-1M dataset and target-modality residual adapters, the model achieves superior fidelity and generalization across 14 seen translation directions and multiple unseen modality combinations.

Localized, High-resolution Geographic Representations with Slepian Functions

This paper constructs a geographic positional encoder using spherical Slepian functions to concentrate representation capacity on a Region of Interest (ROI). It proposes a Slepian-spherical harmonic hybrid encoding to simultaneously account for local high-resolution and global coarse-grained context, consistently outperforming mainstream baselines such as SH, Wavelet, and RFF across five classification, regression, and image enhancement prediction tasks.

The Perception-Physics Paradox: Probing Scientific Alignment with TC-Bench

The authors point out that Vision Foundational Models (VFMs) "seem" proficient at predicting from satellite images but collapse along physical axes in extreme physical regimes. Consequently, this work formalizes the concept of "Scientific Alignment" via "Structural Isomorphism" and releases TC-Bench, a global tropical cyclone benchmark. Through a three-layered suite of linear probes (Static, Dynamic, and Constrained), the authors systematically reveal representation collapse in frozen backbones like DINO, CLIP, SigLIP, and MAE during intense cyclone regimes (\(P_c < 980\) hPa).