Pose Constraints for Consistent Self-supervised Monocular Depth and Ego-Motion

Self-supervised monocular depth estimation approaches suffer not only from scale ambiguity but also infer temporally inconsistent maps w.r.t. scale. While disambiguating during training is possible without some kind of ground truth supervision, having consistent predictions would make it to calculate once inference as a post-processing step and use over-time. With this goal, set temporal consistency losses that minimize pose inconsistencies over time are introduced. Evaluations show introducing these constraints reduces improves the baseline performance ego-motion prediction.