🔍️ Research Interests

Low-level Vision Foundaton Models

• Addressing the limitations of traditional task-specific or dataset-biased methods by building foundation models for low-level vision remote sensing tasks (e.g., denoising, super-resolution) that learn universal representations from large-scale, multimodal visual data, akin to breakthroughs in CV (e.g., CLIP, LLaVA).

Generalizable Low-level Vision

• Developing robust and adaptive algorithms for low-level vision tasks that generalize across diverse scenarios, including varying sensors, imaging conditions, and degradations, etc.

• Harmonizing data-driven deep learning with model-driven priors (e.g., physical degradation models, optimization frameworks) to enhance interpretability, efficiency, and robustness.