Near Real-Time Polarization Rendering

Polarization rendering enables accurate modeling of light-matter interactions but is often computationally intensive. We present a near-real-time polarization ray tracer with indirect illumination on Apple’s Metal GPU framework to generate synthetic indoor polarization images. Generic polarimetric models are often over-parameterized and lack physical insight for common indoor materials, wasting compute and yielding nonphysical interpolations that hinder inverse problems. To address these challenges, we approximate an analytic polarized bidirectional reflectance distribution function (pBRDF) by assuming the non-dominate eigenvalues of the coherency matrix are triply degenerate. This approximation reduces parameter complexity by a factor of two, enforces physicality when interpolated or averaged, and enables importance sampling in the polarimetric domain. On an Apple M4 Max chip, our implementation achieves interactive performance, rendering five-bounce indirect lighting at 200 frames/s and direct-lighting-only scenes at ¿1200 frames/s. This framework serves as a powerful tool for pBRDF visualization and validation, enabling analysis and refinement of existing polarization models.