A Comparative Study of Different CNN Encoders for Monocular Depth Prediction

摘要

Depth estimation of an observed scene is an important task for many domains such as mobile robotics, autonomous driving, and augmented reality. Traditionally, specialized sensors such as stereo cameras and structured light (RGB-D) ones are used to obtain depth along with color information of the environment. However, extending typical monocular cameras with the ability to infer depth information is an attractive solution. In this paper, we will demonstrate a Convolutional Neural Network (CNN) in an encoder-decoder architecture to perform monocular depth prediction. Additionally, we will evaluate and compare different CNN encoders' performance.