We present a novel, GPU-accelerated per-pixel inverse rendering (IR) optimization algorithm based on Particle Swarm Optimization (PSO), IRPSO. IRPSO estimates the per-pixel scene attributes including reflectance properties of a 3D model, and is fast enough to do in situ visualization of the optimization in real-time. We utilize the GPU framebuffer as a computational domain, where each pixel is treated as an independent computational “swarm”. We propose a parallel framework that recovers the reflectance at each pixel location in a massively parallel fashion. The algorithm’s high parallel efficiency is demonstrated through our GPU/GLSL shader implementation of the method. IRPSO is validated experimentally on examples including simulated ground truth images.

Presenter(s)

Koki Nagano , USC Institute for Creative Technologies

Thomas Collins , University of Southern California

Chi-An Chen , University of Southern California

Aiichiro Nakano , University of Southern California