人工智能培训

搜索

人工智能教程:MONET:通过极线发散的多视点半监督关键点(MONET: Multiview Semi-supervised Keypoint via Epip

[复制链接]
baiselong 发表于 2018-6-4 08:30:27 | 显示全部楼层 |阅读模式
baiselong 2018-6-4 08:30:27 649 0 显示全部楼层
人工智能教程:MONET:通过极线发散的多视点半监督关键点(MONET: Multiview Semi-supervised Keypoint via Epipolar Divergence)本文介绍了MONET--一种使用多视图图像流的姿态检测器的端到端半监督学习框架。 MONET与现有模型的区别在于它能够在没有预先训练的模型的情况下检测包括非人类物种在内的一般主题。这样的主要挑战在于专家手动注释的有限可用性,这经常导致检测模型中的大偏差。我们通过以两种方式使用嵌入在未标记数据中的极线约束来解决这个挑战。首先,给定一组标记的数据,可以使用多视图光流在3D中可靠地重建关键点轨迹,导致在时间和空间上几乎不可估量的数据增加详尽的观点。其次,跨视点的检测必须在几何上彼此一致。我们在关键点分布中引入了一种新的几何一致性测量 - 从对极线到对应关键点分布的广义距离。对极分布特征表明当两个视点关键点分布产生零再投影误差时。我们设计atwin网络,通过立体校正可以极大地减轻计算复杂度和训练中的抽样混叠,从而最大限度地减少极线分歧。我们证明我们的框架可以定制不同物种的定制关键点,例如人类,狗和猴子。
This paper presents MONET---an end-to-end semi-supervised learning frameworkfor a pose detector using multiview image streams.What differentiates MONETfrom existing models is its capability of detecting general subjects includingnon-human species without a pre-trained model.A key challenge of such subjectslies in the limited availability of expert manual annotations, which oftenleads to a large bias in the detection model.We address this challenge byusing the epipolar constraint embedded in the unlabeled data in two ways.First, given a set of the labeled data, the keypoint trajectories can bereliably reconstructed in 3D using multiview optical flows, resulting inconsiderable data augmentation in space and time from nearlyexhaustive views.Second, the detection across views must geometrically agree with each other.Weintroduce a new measure of geometric consistency in keypoint distributionscalled epipolar divergence---a generalized distance from the epipolar lines tothe corresponding keypoint distribution.Epipolar divergence characterizes whentwo view keypoint distributions produces zero reprojection error.We design atwin network that minimizes the epipolar divergence through stereorectification that can significantly alleviate computational complexity andsampling aliasing in training.We demonstrate that our framework can localizecustomized keypoints of diverse species, e.g., humans, dogs, and monkeys.人工智能教程:MONET:通过极线发散的多视点半监督关键点(MONET: Multiview Semi-supervised Keypoint via Epipolar Divergence) jDyrDxVm18AzyVAy.jpg
URL地址:https://arxiv.org/abs/1806.00104     ----pdf下载地址:http://arxiv.org/pdf/1806.00104    ----人工智能教程:MONET:通过极线发散的多视点半监督关键点(MONET: Multiview Semi-supervised Keypoint via Epipolar Divergence)
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

baiselong当前离线
新手上路

查看:649 | 回复:0

快速回复 返回顶部 返回列表