Gaussian Process Priors for View-Aware Inference

While frame-independent predictions with deep neural networks have become the prominent solutions to many computer vision tasks, potential benefits of utilizing correlations between frames received less attention. Even though probabilistic machine learning provides ability encode correlation as prior knowledge for inference, there is a tangible gap theory and practice applying methods modern problems. For this, we derive principled framework combine information coupling camera poses (translation orientation) models. We proposed novel view kernel that generalizes standard periodic in SO(3). show how this soft-prior can aid several pose-related tasks like synthesis predict arbitrary points latent space generative models, pointing towards range new applications inter-frame reasoning.