Glare Mitigation using a Differentiable Unified Glare Rating

A new differentiable glare-mitigation pipeline turns the CIE Unified Glare Rating from a hard, discrete evaluation into something gradient-based optimizers can actually use. That matters because glare is a real design constraint in architectural and automotive lighting, but the standard UGR formulation depends on binary thresholds that break smooth inverse rendering workflows.

To make the optimization stable, the method adds a differentiable optical scattering pass that approximates the eye’s point spread function and helps repair fractured evaluation masks under low-sample Monte Carlo conditions. On top of that, the UGR step function is replaced with a tunable sigmoid boundary, so gradients can flow back through the perceptual metric into the physical scene parameters.

The framework is then used across three different control surfaces: roughening surface microgeometry, optimizing boundary-side index of refraction, and masking emitters with gobos. For graphics teams, the practical takeaway is that glare can now be treated as an active objective in physically based pipelines instead of a post-hoc check, which opens the door to more automated comfort-driven lighting design.

The work is slated for the Eurographics Symposium on Rendering 2026 symposium track, and it sits squarely in the growing space where differentiable rendering is being pushed beyond image matching into real design optimization. If this holds up in production-like scenes, it could become a useful tool for technical artists and graphics programmers working on lighting-heavy environments.