Adjoint nonlinear ray tracing

Reconstructing and designing media with continuously-varying refractive index fields remains a challenging problem in computer graphics. A core difficulty trying to tackle this inverse is that light travels inside such along curves, rather than straight lines. Existing techniques for make strong assumptions on the shape of ray medium, thus limit themselves where deflection relatively small. More recently, differentiable rendering have relaxed limitation, by making it possible differentiably simulate curved paths. However, automatic differentiation algorithms underlying these use large amounts memory, restricting existing small low spatial resolutions. We present method optimizing both accounts paths has small, constant memory footprint. adjoint state derive set equations computing derivatives respect field optimization objectives are subject nonlinear tracing constraints. additionally introduce discretization schemes numerically evaluate equations, without need store trajectories significantly reducing requirements our algorithm. technique optimize high-resolution variety applications, including creating different types displays (multiview, lightfield, caustic), gradient-index optics, reconstructing gas flows.