Stokes-Informed Diffusion for Robust Linear Polarization Estimation

2026-07-23

AI Summary

Estimating linear polarization information from a single RGB image is inherently ill-posed, particularly in weakly polarized regions where the angle of polarization (AoP) becomes highly susceptible to noise and unstable. This work proposes GenPolar, a novel framework that, for the first time, integrates Stokes physical priors into a conditional diffusion model. Leveraging the Mueller formalism, GenPolar predicts the Stokes components S₁ and S₂ from the intensity image S₀ and analytically computes the degree of linear polarization (DoLP) and AoP. The method employs an observability-aware loss for supervision, incorporates knowledge distillation for efficient one-step generation, and applies low-rank adaptation (LoRA) to the VAE encoder to mitigate domain-specific autoencoding bias. Evaluated across multiple polarization datasets, GenPolar achieves state-of-the-art performance in DoLP fidelity and AoP stability, significantly enhancing downstream tasks such as material detection and reflection removal.

Abstract

Polarization cues benefit applications such as material detection and de-reflection, yet acquiring them typically requires dedicated hardware. This motivates us to estimate the linear polarization from a single RGB image. However, the task is inherently ill-posed, with the Angle of Polarization (AoP) becoming particularly unstable in weak polarization regions, where the polarimetric signal is overwhelmed by noise, leading to erratic angle estimates. To address these limitations, we propose GenPolar, a Stokes-informed diffusion framework grounded in the Mueller formalism from an intensity observation. Specifically, GenPolar predicts channel-wise linear Stokes components (S1,S2) from intensity S0, from which degree of linear polarization (DoLP) and AoP are analytically derived; AoP is further supervised with an observability-aware loss. In addition, to enable efficient and high-fidelity inference, we adopt a two-stage training strategy. Firstly, a multi-step conditional diffusion model is trained with a physics-based loss. Subsequently, we distill it into a one-step generator, which further supports stable Low-Rank Adaptation (LoRA) of the VAE encoder to mitigate domain-specific autoencoding bias. Extensive experiments across rotating-polarizer, division-of-focal-plane, and hybrid datasets demonstrate that GenPolar achieves state-of-the-art performance in both DoLP fidelity and AoP stability. Crucially, these improvements translate to significant and consistent gains in downstream applications, including material detection and de-reflection.

Problem

• polarization estimation
• Angle of Polarization
• ill-posed problem
• weak polarization
• noise sensitivity

Innovation

• Stokes-informed diffusion
• polarization estimation
• observability-aware loss
• two-stage training
• Low-Rank Adaptation (LoRA)