Abstract
Monocular depth estimation is a foundation task of three-dimensional (3D) reconstruction which is used to improve the accuracy of environment perception. Because of the simpler hardware requirement, it is more suitable than other multi-view methods. In this study, a new monocular depth estimation algorithm based on graph convolution network (GCN) is proposed. The pixel-wise depth relationship is introduced into conventional convolution neural network (CNN) to make up the disadvantage of processing non-Euclidian data. And the remaining depth topological graph information on the spatial latent variables are extracted based on a multi-scale reconstruction strategy. The final results on NYU-v2 depth dataset and KITTI depth dataset demonstrate that our algorithm improves the quality of monocular depth estimation, especially there are several little objects coexisting in the scenes.
Highlights
In the field of image processing, the deep learning networks have achieved a great success in object classification and detection [1]
Fu et al.: Monocular Depth Estimation Based on Multi-Scale graph convolution network (GCN) tasks such as scene detection, scene classification and prediction classification can be significantly improved by the relationship information of objects
A residual neural networks (ResNet)-based model is proposed by [41] which achieves similar performance as ML-GNN does, the decoder layer of former is composed of multi-layer fast-up convolution networks, which is different with the depth regression layer of the convolution neural network (CNN) model
Summary
In the field of image processing, the deep learning networks have achieved a great success in object classification and detection [1]. In monocular depth estimation problem, the deep learning network has a great advantage over the traditional image algorithm. It further reduces the information loss in deep networks than the pooling operation In this way, the model keep a balance between expanding receptive field and maintaining image size. J. Fu et al.: Monocular Depth Estimation Based on Multi-Scale GCNs tasks such as scene detection, scene classification and prediction classification can be significantly improved by the relationship information of objects. They extracts the scene graph from text based description to generate scene images Those mentioned algorithms inspired us that the relationship of pixel-level depth clues have the potential of improving the quality of monocular depth estimation.
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call