PhD student, Tianjin University
2 papers at NeurIPS 2025
MODEM tackles tricky weather image restoration by estimating degradation (DDEM) and using it to guide a novel Morton-order state-space model (MOS2D) for adaptive, context-aware cleaning.
We propose TADiSR, a text-aware diffusion model with joint image super-resolution and segmentation decoders, achieving accurate full-image text SR under real-world degradations.