FUMO: Prior-Modulated Diffusion for Single Image Reflection Removal

arXiv cs.CV / 3/20/2026

📰 NewsTools & Practical UsageModels & Research

共有:

Key Points

FUMO introduces a diffusion-based prior-modulated framework for single image reflection removal, enabling better spatial controllability and structural faithfulness through explicit priors guiding the conditioning.
The method derives two priors from the mixed image: an intensity prior for estimating reflection strength and a high-frequency prior that captures detail-sensitive responses via multi-scale residual aggregation.
A coarse-to-fine training strategy is used, where stage 1 gates conditional residual injections to focus on reflection-dominant and structure-sensitive regions, followed by stage 2 a refinement network that corrects local misalignment and sharpens details.
Experiments on standard benchmarks and challenging real-world images show competitive quantitative results and improved perceptual quality.
The authors release the code on GitHub, supporting reproducibility and practical use.

Abstract

Single image reflection removal (SIRR) is challenging in real scenes, where reflection strength varies spatially and reflection patterns are tightly entangled with transmission structures. This paper presents a diffusion model with prior modulation framework (FUMO) that introduces explicit guidance signals to improve spatial controllability and structural faithfulness. Two priors are extracted directly from the mixed image, an intensity prior that estimates spatial reflection severity and a high-frequency prior that captures detail-sensitive responses via multi-scale residual aggregation. We propose a coarse-to-fine training paradigm. In the first stage, these cues are combined to gate the conditional residual injections, focusing the conditioning on regions that are both reflection-dominant and structure-sensitive. In the second stage, a fine-grained refinement network corrects local misalignment and sharpens fine details in the image space. Experiments conducted on both standard benchmarks and challenging images in the wild demonstrate competitive quantitative results and consistently improved perceptual quality. The code is released at https://github.com/Lucious-Desmon/FUMO.