一种基于U型全卷积神经网络深度估计模型.pdfVIP

Computer Science and Application 计算机科学与应用, 2019, 9(2), 250-255 Published Online February 2019 in Hans. /journal/csa /10.12677/csa.2019.92029 Image Depth Estimation Model Based on Fully Convolutional U-Net Xiaokang Wang, Xiaoning Fu School of Electromechanical Engineering, Xidian University, Xi’an Shaanxi th th st Received: Jan . 15 , 2018; accepted: Jan. 25 , 2019; published: Feb. 1 , 2019 Abstract The problem of depth estimation from single image has been addressed. The mapping between a single image and the depth map is inherently ambiguous, and requires both global and local in- formation. This paper presents a fully convolutional U-net whose encoder is pre-trained ResNet50 without fully connected layer or pooling layer, and uses residual up-sampling layers to enlarge the feature maps. Besides, skip connection is introduced, making the model U-net, to fuse global and local information. The network can be end-to-end trained. Keywords Depth Estimation, Fully Convolutional Network, Residual Up-Sampling Layers, Skip Connection 一种基于U型全卷积神经网络的深度估计模型 王小康，付小宁 西安电子科技大学机电工程学院，陕西 西安 收稿日期：2019年1月15 日；录用日期：2019年1月25 日；发布日期：2019年2月1 日 摘 要 本文解决了从单张图像估计深度信息的问题。单张图像与深度图之间的映射是是模棱两可的，它需要全 局信息和局部信息。本文部署了一个全卷积U型神经网络，它用预训练的ResNet-50网络提取图像特征， 然后用残差上采样模块将特征图恢复到深度图的尺寸大小，并且使用了跳跃链接，整个网络呈现U型， 从而对全局信息和局部信息进行融合。整个网络可以进行端到端的训练。 文章引用: 王小康, 付小宁. 一种基于 U 型全卷积神经网络的深度估计模型[J]. 计算机科学与应用, 2019, 9(2): 250-255. DOI: 10.12677/csa.2019.92029 王小康，付小宁 关键词 单目深度估计，全卷积神经网络，残差上采样，跳跃链接 Copyright © 2019 by author(s) and Hans Publishers Inc. This work is licensed under the Creative Commons Attribution International License (CC BY). /licenses/by