Bayesian Deep Neural Networks for Supervised Learning of Single-View Depth
This work addresses uncertainty estimation for robotic perception to prevent collisions, but it is incremental as it builds on existing methods like MC dropout and deep ensembles.
The paper tackled uncertainty quantification in single-view supervised depth learning by evaluating MC dropout and deep ensembles, showing that dropout in all encoder layers performs similarly to deep ensembles with lower memory usage and demonstrating its application in pseudo-RGBD ICP for accurate two-view relative motion estimation.
Uncertainty quantification is essential for robotic perception, as overconfident or point estimators can lead to collisions and damages to the environment and the robot. In this paper, we evaluate scalable approaches to uncertainty quantification in single-view supervised depth learning, specifically MC dropout and deep ensembles. For MC dropout, in particular, we explore the effect of the dropout at different levels in the architecture. We show that adding dropout in all layers of the encoder brings better results than other variations found in the literature. This configuration performs similarly to deep ensembles with a much lower memory footprint, which is relevant forapplications. Finally, we explore the use of depth uncertainty for pseudo-RGBD ICP and demonstrate its potential to estimate accurate two-view relative motion with the real scale.