Multi-view stereo by temporal nonparametric fusion

Video material for the paper 'Multi-view stereo by temporal nonparametric fusion' by Yuxin Hou, Juho Kannala, and Arno Solin. arXiv: arxiv.org/abs/1904.06397
We propose a method for depth estimation from multiple monocular camera views that are tied together by a pose-aware Gaussian process prior. This method is capable of running in real-time on a tablet/phone (the video shows how it works on an iPad). The training data was not captured on a similar device, nor environment.
More information available on the project page: aaltoml.github.io/GP-MVS/