General Depth Estimation For Machine Vision Using Sparse Depths

The ubiquity of single camera systems in society has made improving monocular depth estimation a topic of increasing interest in the broader computer vision community. Inspired by recent work that has shown the effectiveness of utilizing sparse samples in a depth estimation pipeline, researchers aim to augment performance by enforcing additional constraints governed by the predictions of Simultaneous Localization And Mapping (SLAM) algorithms. Using these constraints as a discriminator to isolate depth predictions that deviate the most from the norm, researchers aim to show that improving performance on these select depth predictions will lead to better depth prediction overall.