人工智能培训

搜索

机器学习论文:物理原始分解(Physical Primitive Decomposition)

[复制链接]
rrrrrrr 发表于 2018-9-14 08:58:45 | 显示全部楼层 |阅读模式
rrrrrrr 2018-9-14 08:58:45 46 0 显示全部楼层
机器学习论文:物理原始分解(Physical Primitive Decomposition)物体由零件组成,每个零件具有不同的几何形状,物理,功能和可供性。开发这种分布式,物理的,可解释的对象表示将有助于智能代理更好地探索和与世界交互。在本文中,我们研究了物理原型分解 - 通过其组件理解一个对象,每个组件都有物理和几何属性。由于对象部分和物理学的注释数据很少,我们提出了一种新的公式,通过解释物体的外观及其在物理事件中的行为来学习物理原型。我们的模型在合理的和真实的场景中都能在块塔和工具上表现良好;我们还证明视觉和物理观测通常提供互补信号。我们进一步展示消融和行为研究,以更好地理解我们的模型并将其与人类表现进行对比。
Objects are made of parts, each with distinct geometry, physics,functionality, and affordances.Developing such a distributed, physical,interpretable representation of objects will facilitate intelligent agents tobetter explore and interact with the world.In this paper, we study physicalprimitive decomposition---understanding an object through its components, eachwith physical and geometric attributes.As annotated data for object parts andphysics are rare, we propose a novel formulation that learns physicalprimitives by explaining both an object's appearance and its behaviors inphysical events.Our model performs well on block towers and tools in bothsynthetic and real scenarios;we also demonstrate that visual and physicalobservations often provide complementary signals.We further present ablationand behavioral studies to better understand our model and contrast it withhuman performance.机器学习论文:物理原始分解(Physical Primitive Decomposition) v5DoV51OZ1Gz0VpO.jpg
URL地址:https://arxiv.org/abs/1809.05070     ----pdf下载地址:https://arxiv.org/pdf/1809.05070    ----机器学习论文:物理原始分解(Physical Primitive Decomposition)
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

rrrrrrr当前离线
新手上路

查看:46 | 回复:0

快速回复 返回顶部 返回列表