人工智能培训

搜索

机器学习培训:STEREONET:用于实时边缘感知深度预测的引导分层细化(StereoNet: Guided Hierarchical Refinement f

[复制链接]
jamiezhao 发表于 2018-7-25 09:56:42 | 显示全部楼层 |阅读模式
jamiezhao 2018-7-25 09:56:42 1808 0 显示全部楼层
机器学习培训:STEREONET:用于实时边缘感知深度预测的引导分层细化(StereoNet: Guided Hierarchical Refinement for Real-Time Edge-Aware Depth  Prediction)本文介绍了StereoNet,这是第一个用于实时立体匹配的端到端深度架构,在NVidia Titan X上以60 fps运行,可生成高质量,边缘保留,无量化的视差图。本文的一个重要见解是网络实现了亚像素匹配精度,而不是传统立体匹配方法的高度。这使我们能够通过使用非常低分辨率的成本体积来实现实时性能,该体积可以编码所需的所有信息实现高差距精度。通过采用学习的边缘识别上采样功能来实现空间精度。我们的模型使用Siamese网络从左右图像中提取特征。在非常低分辨率的成本体积中计算视差的第一估计,然后分层地通过使用紧凑的像素到像素细化网络的学习的上采样函数来重新引入高频细节。利用颜色输入作为指导,该功能能够产生高质量的边缘感知输出。我们在多个基准测试中获得了令人瞩目的成果,展示了所提出的方法如何在可接受的计算预算下提供极大的灵
This paper presents StereoNet, the first end-to-end deep architecture forreal-time stereo matching that runs at 60 fps on an NVidia Titan X, producinghigh-quality, edge-preserved, quantization-free disparity maps.A key insightof this paper is that the network achieves a sub-pixel matching precision thanis a magnitude higher than those of traditional stereo matching approaches.This allows us to achieve real-time performance by using a very low resolutioncost volume that encodes all the information neededto achieve high disparityprecision.Spatial precision is achieved by employing a learned edge-awareupsampling function.Our model uses a Siamese network to extract features fromthe left and right image.A first estimate of the disparity is computed in avery low resolution cost volume, then hierarchically the model re-introduceshigh-frequency details through a learned upsampling function that uses compactpixel-to-pixel refinement networks.Leveraging color input as a guide, thisfunction is capable of producing high-quality edge-aware output.We achievecompelling results on multiple benchmarks, showing how the proposed methodoffers extreme flexibility at an acceptable computational budget.机器学习培训:STEREONET:用于实时边缘感知深度预测的引导分层细化(StereoNet: Guided Hierarchical Refinement for Real-Time Edge-Aware Depth  Prediction) IyZivGEnCyeBEYFv.jpg
URL地址:https://arxiv.org/abs/1807.08865     ----pdf下载地址:https://arxiv.org/pdf/1807.08865    ----机器学习培训:STEREONET:用于实时边缘感知深度预测的引导分层细化(StereoNet: Guided Hierarchical Refinement for Real-Time Edge-Aware Depth  Prediction)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

jamiezhao当前离线
新手上路

查看:1808 | 回复:0

快速回复 返回顶部 返回列表