Research map / Image generation / Image restoration
Learning to Refine: Spectral-Decoupled Iterative Refinement Framework for Precipitation Nowcasting
Yunlong Zhou, Chen Zhao, Danyang Peng, Fanfan Ji, Xiaotong Yuan
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add FR-Refiner to SFG-Former (ablation) | SDIR · Shanghai | HSS avg: 0.3529 -> 0.5367 | Improved |
| add SFG-Former to FR-Refiner (ablation) | SDIR · Shanghai | HSS avg: 0.4614 -> 0.5367 | Improved |
| add PCPSD loss (ablation) | SDIR · Shanghai | HSS avg: 0.5367 -> 0.5882 | Improved |
| use Beta-distributed instead of uniform sampling of s (ablation) | SDIR · Shanghai | HSS avg: 0.2842 -> 0.5882 | Improved |
| add adaptive normalization (AN) (ablation) | SDIR · Shanghai | HSS avg: 0.5073 -> 0.5882 | Improved |
| use spectral curriculum loss instead of pure MAE loss (ablation) | SDIR · Shanghai | HSS avg: HSS 0.4509 -> 0.5882; MAE 1106.4 -> 1129.1 | Mixed |
| use 8 inference steps instead of 1 (ablation) | SDIR · Shanghai | HSS avg: 0.5584 -> 0.5882 | Improved |
| use 16 inference steps instead of 8 (ablation) | SDIR · Shanghai | HSS avg: HSS 0.5882 -> 0.5898; MAE 1129.1 -> 1190.1 | Mixed |
| use 32 inference steps instead of 8 (ablation) | SDIR · Shanghai | HSS avg: 0.5882 -> 0.5564 | Hurt |
| increase Beta skew from Beta(1.0,3.0) to Beta(1.0,3.5) (ablation) | SDIR · Shanghai | HSS avg: 0.5882 -> 0.5582 | Hurt |
| use Beta(1.0,2.5) instead of Beta(1.0,3.0) (ablation) | SDIR · Shanghai | HSS avg: 0.5882 -> 0.4835 | Hurt |
| raise PCPSD weight eta from 0.01 to 0.1 (ablation) | SDIR · Shanghai | HSS avg: 0.5882 -> 0.5405 | Hurt |
Closest papers by meaning
- PostCast: Generalizable Postprocessing for Precipitation Nowcasting via Unsupervised Blurriness Modeling (ICLR 2025 · similarity 0.88)
- Fourier Amplitude and Correlation Loss: Beyond Using L2 Loss for Skillful Precipitation Nowcasting (NeurIPS 2024 · similarity 0.85)
- Perceptually Constrained Precipitation Nowcasting Model (ICML 2025 · similarity 0.85)
- Scaling Storm-Resolving Atmospheric AI Simulation to the Entire Planet (NeurIPS 2026 · similarity 0.85)
- Precipitation Downscaling with Spatiotemporal Video Diffusion (NeurIPS 2024 · similarity 0.85)
- CasCast: Skillful High-resolution Precipitation Nowcasting via Cascaded Modelling (ICML 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
