Lightweight Monocular Depth with a Novel Neural Architecture Search Method

Abstract

This paper presents a novel neural architecture search method, called LiDNAS, for generating lightweight monocular depth estimation models. Unlike previous neural architecture search (NAS) approaches, where finding optimized networks is computationally demanding, the introduced novel Assisted Tabu Search leads to efficient architecture exploration. Moreover, we construct the search space on a pre-defined backbone network to balance layer diversity and search space size. The LiDNAS method outperforms the state-of-the-art NAS approach, proposed for disparity and depth estimation, in terms of search efficiency and output model performance. The LiDNAS optimized models achieve result superior to compact depth estimation state-of-the-art on NYU-Depth-v2, KITTI, and Scan-Net, while being 7%-500% more compact in size, i.e the number of model parameters.

Quadtree Segmentation Network for Obstacle Avoidance in Monocular Navigation

“…The results obtained for monocular depth prediction are encouraging and raise the question of the use of these methods for navigation applications as expressed in Dong et al survey [3]. For the moment, only a minority of papers address the issue of lightweight architecture, usable on embedded systems [14], [10]. These methods seek to optimize their network structure to find the best trade-off between lightness and precision.…”

“…However, most of these methods are engaged in a race for accuracy [13] at the expense of computation time. Only a few of them addressed the question of real-time embedded solutions [14], [10].…”