Geo-Supervised Visual Depth Prediction
ICRA• 2019
Abstract
We propose using global orientation from inertial measurements, and the bias
it induces on the shape of objects populating the scene, to inform visual 3D
reconstruction. We test the effect of using the resulting prior in depth
prediction from a single image, where the normal vectors to surfaces of objects
of certain classes tend to align with gravity or be orthogonal to it. Adding
such a prior to baseline methods for monocular depth prediction yields
improvements beyond the state-of-the-art and illustrates the power of gravity
as a supervisory signal.