Perceptual coding for 3D reconstruction

A primary issue in 3D reconstruction is the realtime efficacy of different coding methods for the multiple decisions among competing 3D solutions. A common model framework making such coding decisions is the boundary limited drift-diffusion model, which has been developed in parallel in various branches of science from quantum physics to economics. A common property of all such models is the linear increase in variance of the diffusion processes over time, implying an inability to focus on the current information in the environment, and the inevitability of a forced random decision in the absence of any reliable evidence. We have developed an alternative, more plausible model framework for Bayesian information accumulation that solves both problems and provides an accurate account of many features of context effects in human 3D reconstruction performance.