Learning a Depth Covariance Function
This work addresses geometric vision problems for computer vision researchers, but it appears incremental as it builds on existing covariance function techniques.
The paper tackles the problem of geometric vision tasks by proposing a learned depth covariance function that can define priors over depth functions, predictive distributions, and active point selection methods, with applications to depth completion, bundle adjustment, and monocular dense visual odometry, but no concrete numbers are provided for results.
We propose learning a depth covariance function with applications to geometric vision tasks. Given RGB images as input, the covariance function can be flexibly used to define priors over depth functions, predictive distributions given observations, and methods for active point selection. We leverage these techniques for a selection of downstream tasks: depth completion, bundle adjustment, and monocular dense visual odometry.