Geo-Supervised Visual Depth Prediction
This work addresses depth estimation for computer vision applications, offering a domain-specific enhancement.
The paper tackled the problem of monocular depth prediction by incorporating a gravity prior from inertial measurements to bias object surface normals, resulting in improvements beyond state-of-the-art methods.
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.