人工智能培训

搜索

深度学习论文:多人运动捕捉的神经场景分解(Neural Scene Decomposition for Multi-Person Motion Capture)

[复制链接]
einter 发表于 2019-3-15 12:24:46 | 显示全部楼层 |阅读模式
einter 2019-3-15 12:24:46 430 0 显示全部楼层
深度学习论文:多人运动捕捉的神经场景分解(Neural Scene Decomposition for Multi-Person Motion Capture)学习一般图像表示已被证明是许多计算机视觉任务成功的关键。例如,图像理解问题的许多方法依赖于最初在ImageNet上训练的深度网络,主要是因为学习的特征是从有限标记数据中学习的有价值的起点。然而,当谈到多人的3D动作捕捉时,这些特征的用途有限。因此,在本文中,我们提出了一种学习可用于此目的的特征的方法。为此,我们引入了一种自我监督的方法来学习我们称之为神经场景分解(NSD)的神经场景分解(NSD),可用于3D姿态估计。 NSD包括三层抽象来表示人类主体:空间布局的边界框和相对深度;根据实例分割掩码的2D形状表示;和特定主题的外观和3D姿势信息。通过利用来自多视图数据的自我监督,我们的NSD模型可以在没有任何2D或3D监督的情况下从头到尾进行。与之前的方法相比,它适用于多人和全帧图像。因为它可以对3D几何进行编码,所以可以有效地利用NSD从少量注释数据中训练3D姿态估计网络。
Learning general image representations has proven key to the success of manycomputer vision tasks.For example, many approaches to image understandingproblems rely on deep networks that were initially trained on ImageNet, mostlybecause the learned features are a valuable starting point to learn fromlimited labeled data.However, when it comes to 3D motion capture of multiplepeople, these features are only of limited use.In this paper, we therefore propose an approach to learning features that areuseful for this purpose.To this end, we introduce a self-supervised approachto learning what we call a neural scene decomposition (NSD) that can beexploited for 3D pose estimation.NSD comprises three layers of abstraction torepresent human subjects: spatial layout in terms of bounding-boxes andrelative depth;a 2D shape representation in terms of an instance segmentationmask;and subject-specific appearance and 3D pose information.By exploitingself-supervision coming from multiview data, our NSD model can be trainedend-to-end without any 2D or 3D supervision.In contrast to previousapproaches, it works for multiple persons and full-frame images.Because itencodes 3D geometry, NSD can then be effectively leveraged to train a 3D poseestimation network from small amounts of annotated data.深度学习论文:多人运动捕捉的神经场景分解(Neural Scene Decomposition for Multi-Person Motion Capture) SUJoO5sH07V5BUuB.jpg
URL地址:https://arxiv.org/abs/1903.05684     ----pdf下载地址:https://arxiv.org/pdf/1903.05684    ----深度学习论文:多人运动捕捉的神经场景分解(Neural Scene Decomposition for Multi-Person Motion Capture)
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

einter当前离线
新手上路

查看:430 | 回复:0

快速回复 返回顶部 返回列表