English

InvFusion: Bridging Supervised and Zero-shot Diffusion for Inverse Problems

Computer Vision and Pattern Recognition 2025-11-20 v2

Abstract

Diffusion Models have demonstrated remarkable capabilities in handling inverse problems, offering high-quality posterior-sampling-based solutions. Despite significant advances, a fundamental trade-off persists regarding the way the conditioned synthesis is employed: Zero-shot approaches can accommodate any linear degradation but rely on approximations that reduce accuracy. In contrast, training-based methods model the posterior correctly, but cannot adapt to the degradation at test-time. Here we introduce InvFusion, the first training-based degradation-aware posterior sampler. InvFusion combines the best of both worlds -- the strong performance of supervised approaches and the flexibility of zero-shot methods. This is achieved through a novel architectural design that seamlessly integrates the degradation operator directly into the diffusion denoiser. We compare InvFusion against existing general-purpose posterior samplers, both degradation-aware zero-shot techniques and blind training-based methods. Experiments on the FFHQ and ImageNet datasets demonstrate state-of-the-art performance. Beyond posterior sampling, we further demonstrate the applicability of our architecture, operating as a general Minimum Mean Square Error predictor, and as a Neural Posterior Principal Component estimator.

Keywords

Cite

@article{arxiv.2504.01689,
  title  = {InvFusion: Bridging Supervised and Zero-shot Diffusion for Inverse Problems},
  author = {Noam Elata and Hyungjin Chung and Jong Chul Ye and Tomer Michaeli and Michael Elad},
  journal= {arXiv preprint arXiv:2504.01689},
  year   = {2025}
}
R2 v1 2026-06-28T22:43:50.685Z