Research map / Image generation / Super-resolution
Breaking Scale Anchoring: Frequency Representation Learning for Accurate High-Resolution Inference from Low-Resolution Training
Wenshuo Wang, Fan Zhang
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 Frequency Representation Learning (multi-resolution training, Nyquist-normalized frequency encoding, frequency consistency loss) | seven architectures (GNN, Transformer, CNN, Diffusion, NO, Neural ODE, NN) · 3D HIT fluid simulation, 129^3 | RMSE: 3.57x-7.74x lower RMSE; RMSE ratio 1.017-1.338 -> 0.135-0.375 | Improved |
| add FRL to WeatherGFT Transformer | WeatherGFT · ERA5 Z500 7-day, 1440x721 | ACC: 0.44 -> 0.65 | Improved |
| add FRL to ClimODE | ClimODE · ERA5 Z500 7-day, 1440x721 | RMSE: 772 -> 502 | Improved |
| increase test resolution without FRL | baseline STF models · 3D HIT fluid simulation | RMSE ratio: 1.017-1.338 (error stays anchored) | No clear effect |
| increase test resolution beyond training Nyquist (ablation) | baseline models on 2D convection-diffusion · Error Ratio | bandlimited/wideband error ratio: 1.000 (64x64) -> 0.223 (GNN, 256x256) | Hurt |
| FRL training cost overhead | FRL variants · 3D fluid simulation | training time / memory: 1.1x-1.4x time, 1.3x-1.5x memory, <2% inference overhead | Hurt |
Closest papers by meaning
- Pushing the Limits of High-Resolution Weather Forecasting through Data Scaling (ECCV 2026 · similarity 0.84)
- Frequency-Aware Autoregressive Modeling for Efficient High-Resolution Image Synthesis (ICCV 2025 · similarity 0.84)
- The False Promise of Zero-Shot Super-Resolution in Machine-Learned Operators (ICLR 2026 · similarity 0.84)
- Hierarchical Neural Operator Transformer with Learnable Frequency-aware Loss Prior for Arbitrary-scale Super-resolution (ICML 2024 · similarity 0.84)
- Everything at Every Scale: Scale-Invariant Diffusion with Continuous Super-Resolution (NeurIPS 2026 · similarity 0.83)
- FreSh: Frequency Shifting for Accelerated Neural Representation Learning (ICLR 2025 · similarity 0.83)
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.
