PolarAnything: Diffusion-based Polarimetric Image Synthesis

TL;DR

PolarAnything is a diffusion-based model that synthesizes photorealistic polarization images from a single RGB image, overcoming limitations of existing simulators and enabling applications like 3D shape reconstruction.

Contribution

It introduces a novel diffusion framework for polarization image synthesis that does not require extensive 3D assets, improving accessibility and realism.

Findings

Generates high-quality polarization images from a single RGB input

Supports downstream tasks such as shape from polarization

Outperforms existing polarization simulators in realism and accuracy

Abstract

Polarization images facilitate image enhancement and 3D reconstruction tasks, but the limited accessibility of polarization cameras hinders their broader application. This gap drives the need for synthesizing photorealistic polarization images. The existing polarization simulator Mitsuba relies on a parametric polarization image formation model and requires extensive 3D assets covering shape and PBR materials, preventing it from generating large-scale photorealistic images. To address this problem, we propose PolarAnything, capable of synthesizing polarization images from a single RGB input with both photorealism and physical accuracy, eliminating the dependency on 3D asset collections. Drawing inspiration from the zero-shot performance of pretrained diffusion models, we introduce a diffusion-based generative framework with an effective representation strategy that preserves the fidelity of polarization properties. Experiments show that our model generates high-quality polarization images and supports downstream tasks like shape from polarization.