Paper Detail
Sedjro Salomon Hotegni, Konstantin Sonntag, Paul-Christian Bürkner, Sebastian Peitz
Flow-matching models can produce high-quality image restorations, but their learned dynamics may amplify small input perturbations and make reconstructions unstable. We introduce LipsFlower to reduce this amplification. We derive stability bounds for Flow-from-Y, which initializes the ordinary differential equation from a noise-perturbed observation, as well as for observation-conditioned posterior sampling, which conditions the velocity field on the observation throughout integration. Based on this analysis, we develop LipHDiT, a Lipschitz-aware hourglass diffusion transformer that serves as the vector-field architecture of LipsFlower. LipHDiT incorporates stabilized approximate spectral normalization, bounded modulation, and convex residual mixing. We also use adversarial finite-radius gain regularization at sampled interpolation states to penalize excessive sensitivity to the state and, for conditional sampling, the observation. We test both formulations on blind face restoration, measuring adversarial robustness and the perceptual quality of the reconstructions. LipsFlower reduces perturbation amplification while keeping restoration quality competitive with the evaluated baselines. These results suggest that Lipschitz-aware flow matching can improve the stability of generative image restoration.
No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.
No ranking explanation is available yet.
No tags.
@misc{hotegni2026lipschitz,
title = {Lipschitz-Controlled Flow Matching for Robust Image Restoration},
author = {Sedjro Salomon Hotegni and Konstantin Sonntag and Paul-Christian Bürkner and Sebastian Peitz},
year = {2026},
abstract = {Flow-matching models can produce high-quality image restorations, but their learned dynamics may amplify small input perturbations and make reconstructions unstable. We introduce LipsFlower to reduce this amplification. We derive stability bounds for Flow-from-Y, which initializes the ordinary differential equation from a noise-perturbed observation, as well as for observation-conditioned posterior sampling, which conditions the velocity field on the observation throughout integration. Based on },
url = {https://huggingface.co/papers/2611.13995},
keywords = {flow-matching models, image restoration, learned dynamics, input perturbations, Flow-from-Y, ordinary differential equation, noise-perturbed observation, observation-conditioned posterior sampling, velocity field, LipsFlower, LipHDiT, Lipschitz-aware hourglass diffusion transformer, vector-field architecture, stabilized approximate spectral normalization, bounded modulation, convex residual mixing, adversarial finite-radius gain regularization, interpolation states, conditional sampling, blind face restoration, adversarial robustness, perceptual quality, perturbation amplification, Lipschitz-aware flow matching, generative image restoration, huggingface daily},
eprint = {2611.13995},
archiveprefix = {arXiv},
}
{}