CaltechAUTHORS
  A Caltech Library Service

Fast Uncertainty Estimation for Deep Learning Based Optical Flow

Lee, Serin and Capuano, Vincenzo and Harvard, Alexei and Chung, Soon-Jo (2020) Fast Uncertainty Estimation for Deep Learning Based Optical Flow. . (Unpublished) https://resolver.caltech.edu/CaltechAUTHORS:20200805-121745448

[img] PDF - Accepted Version
See Usage Policy.

1193Kb

Use this Persistent URL to link to this item: https://resolver.caltech.edu/CaltechAUTHORS:20200805-121745448

Abstract

We present a novel approach to reduce the processing time required to derive the estimation uncertainty map in deep learning-based optical flow determination methods. Without uncertainty aware reasoning, the optical flow model, especially when it is used for mission critical fields such as robotics and aerospace, can cause catastrophic failures. Although several approaches such as the ones based on Bayesian neural networks have been proposed to handle this issue, they are computationally expensive. Thus, to speed up the processing time, our approach applies a generative model, which is trained by input images and an uncertainty map derived through a Bayesian approach. By using synthetically generated images of spacecraft, we demonstrate that the trained generative model can produce the uncertainty map 100~700 times faster than the conventional uncertainty estimation method used for training the generative model itself. We also show that the quality of uncertainty map derived by the generative model is close to that of the original uncertainty map. By applying the proposed approach, the deep learning model operated in real-time can avoid disastrous failures by considering the uncertainty as well as achieving better performance removing uncertain portions of the prediction result.


Item Type:Report or Paper (Discussion Paper)
ORCID:
AuthorORCID
Capuano, Vincenzo0000-0002-6886-5719
Chung, Soon-Jo0000-0002-6657-3907
Additional Information:The authors thank A. Rahmani, A. Santamaria-Navarro, and F. Y. Hadaegh for their technical guidance.
Group:GALCIT
Record Number:CaltechAUTHORS:20200805-121745448
Persistent URL:https://resolver.caltech.edu/CaltechAUTHORS:20200805-121745448
Usage Policy:No commercial reproduction, distribution, display or performance rights in this work are provided.
ID Code:104758
Collection:CaltechAUTHORS
Deposited By: Tony Diaz
Deposited On:05 Aug 2020 19:28
Last Modified:05 Aug 2020 19:28

Repository Staff Only: item control page