[ASASR] Coloring the Noise: Adversarial Sobolev Alignment for Faithful Image Super-Resolution

Official repository for the ICML 2026 paper:

Coloring the Noise: Adversarial Sobolev Alignment for Faithful Image Super-Resolution Hongbo Wang, Huaibo Huang, Pin Wang, Jinhua Hao, Chao Zhou, Ran He International Conference on Machine Learning (ICML), 2026 [arXiv:2605.23264]

---

Highlights

• 🎨 Colored-Noise Diffusion — replaces isotropic Gaussian noise with a spectrally shaped kernel aligned to the natural image manifold.
• 📐 Sobolev-Induced Geometry — reformulates the generative flow under a Riemannian metric that respects high-frequency structure.
• ⚔️ Adversarial Sobolev Alignment — a Riesz-representation-based adversary produces worst-case negatives for preference optimization.
• 🖼️ Faithful Super-Resolution — strong improvements in spectral consistency, structural fidelity, and artifact suppression.

---

🚧 Code Release

We are actively cleaning up the codebase. The code will be open-sourced in this repository before ICML 2026 (July 2026) — well ahead of the conference.

Please star ⭐ and watch 👀 this repo to be notified the moment the release lands.

Citation

If you find our work useful, please consider citing:

@inproceedings{wang2026asasr,
  title     = {Coloring the Noise: Adversarial Sobolev Alignment for Faithful Image Super-Resolution},
  author    = {Wang, Hongbo and Huang, Huaibo and Wang, Pin and Hao, Jinhua and Zhou, Chao and He, Ran},
  booktitle = {International Conference on Machine Learning (ICML)},
  year      = {2026}
}