🌍 Earth Science

🤖 AAAI2026 · 1 paper notes

MdaIF: Robust One-Stop Multi-Degradation-Aware Image Fusion with Language-Driven Semantics

This paper proposes MdaIF, a framework that leverages a vision-language model (VLM) to extract degradation-aware semantic priors for guiding mixture-of-experts (MoE) routing and channel attention modulation, enabling one-stop infrared-visible image fusion across multiple degradation scenarios without requiring degradation-type annotations.