Research map / Image generation / Super-resolution
PODiff: Latent Diffusion in Proper Orthogonal Decomposition Space for Scientific Super-Resolution
Onkar Jadhav, Tim French, Matthew Rayson, Nicole L. Jones
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 |
|---|---|---|---|
| use POD basis instead of random orthonormal basis (ablation) | PODiff-K40 · SST downscaling WA coast (2011 test) | RMSE (C): 0.9987 -> 0.3923 | Improved |
| add diffusion on top of POD projection (ablation) | POD-proj · SST downscaling WA coast (2011 test) | RMSE (C): 0.7084 -> 0.3923 | Improved |
| increase POD latent dimension K from 10 to 20 (ablation) | PODiff · SST downscaling WA coast (2011 test) | RMSE (C): 0.7725 -> 0.5171 | Improved |
| increase POD latent dimension K from 20 to 40 (ablation) | PODiff · SST downscaling WA coast (2011 test) | RMSE (C): 0.5171 -> 0.3923 | Improved |
| diffuse in POD space instead of pixel space | PixelDiff · SST downscaling WA coast (2011 test) | RMSE (C): 0.4118 -> 0.3923 | Improved |
| diffuse in POD space instead of pixel space | PixelDiff · SST downscaling WA coast (2011 test) | peak GPU memory (GB): 12.5 GB -> 1.4 GB | Improved |
| diffuse in POD space instead of pixel space | PixelDiff · SST downscaling WA coast (2011 test) | training time: 48 h -> 3.8 h | Improved |
| use PODiff instead of MC Dropout U-Net for uncertainty | MC Dropout U-Net · SST downscaling WA coast (2011 test), 20 test days | CRPS: 0.4821 -> 0.2889 | Improved |
| reduce U-Net capacity (C=128 to C=32) (ablation) | U-Net · SST downscaling WA coast (2011 test) | RMSE (C): 0.6788 -> 0.6819 | No clear effect |
| increase ensemble size from M=50 to M=100 (ablation) | PODiff-K40 · SST downscaling WA coast (2011 test), 90% coverage | empirical coverage: 0.899 -> 0.901 | No clear effect |
| increase ensemble size from M=100 to M=200 (ablation) | PODiff-K40 · SST downscaling WA coast (2011 test), 90% coverage | empirical coverage: 0.901 -> 0.902 | No clear effect |
Closest papers by meaning
- Generating Full-field Evolution of Physical Dynamics from Irregular Sparse Observations (NeurIPS 2025 · similarity 0.86)
- BUFF: Bayesian Uncertainty Guided Diffusion Probabilistic Model for Single Image Super-Resolution (AAAI 2025 · similarity 0.86)
- Guided Diffusion Sampling on Function Spaces with Applications to PDEs (NeurIPS 2025 · similarity 0.85)
- Tortoise and Hare Guidance: Accelerating Diffusion Model Inference with Multirate Integration (NeurIPS 2025 · similarity 0.85)
- A Novel Diffusion Model for Pairwise Geoscience Data Generation with Unbalanced Training Dataset (AAAI 2025 · similarity 0.85)
- DiffATS: Diffusion in Aligned Tensor Space (NeurIPS 2026 · similarity 0.85)
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.
