Bayesian Deep Neural Networks for Supervised Learning of Single-View Depth

Rodriguez-Puigvert, J. ; Martinez-Cantin, R. (Universidad de Zaragoza) ; Civera Sancho, J. (Universidad de Zaragoza)

Resumen: 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 letter, 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 for applications. 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. 2377-3766 © 2022 IEEE.

Idioma: Inglés

DOI: 10.1109/LRA.2022.3142915

Año: 2022

Publicado en: IEEE Robotics and Automation Letters 7, 2 (2022), 2565-2572
ISSN: 2377-3766

Factor impacto JCR: 5.2 (2022)

Categ. JCR: ROBOTICS rank: 10 / 30 = 0.333 (2022) - Q2 - T2

Factor impacto CITESCORE: 7.6 - Engineering (Q1) - Mathematics (Q1) - Computer Science (Q1)

Factor impacto SCIMAGO: 1.693 - Artificial Intelligence (Q1) - Biomedical Engineering (Q1) - Computer Science Applications (Q1) - Mechanical Engineering (Q1) - Control and Optimization (Q1) - Control and Systems Engineering (Q1) - Human-Computer Interaction (Q1) - Computer Vision and Pattern Recognition (Q1)

Tipo y forma: Artículo (Versión definitiva)

Área (Departamento): Área Ingen.Sistemas y Automát. (Dpto. Informát.Ingenie.Sistms.)