人工智能培训

搜索

机器学习论文:RVOS:用于视频对象分割的端到端循环网络(RVOS: End-to-End Recurrent Network for Video Object

[复制链接]
kkkgame 发表于 2019-3-15 13:07:05 | 显示全部楼层 |阅读模式
kkkgame 2019-3-15 13:07:05 236 0 显示全部楼层
机器学习论文:RVOS:用于视频对象分割的端到端循环网络(RVOS: End-to-End Recurrent Network for Video Object Segmentation)多目标视频对象分割是一项具有挑战性的任务,特别是对于零射击情况,当在初始帧处没有给出对象掩模并且模型必须找到沿着序列分割的对象时。在我们的工作中,我们提出了一个可以完全端到端训练的多目标视频对象分割(RVOS)的循环网络。我们的模型在两个不同的域上复制:(i)空间,它允许发现帧内的不同对象实例,以及(ii)时间,以保持分段对象沿时间的一致性。我们为零射击视频对象分割训练RVOS,并且是第一个报告DAVIS-2017和YouTube-VOS基准的定量结果的人。此外,通过使用在先前时间步骤中获得的掩模作为由循环模块处理的输入,我们使用RVOS进行单次视频对象分割。我们的模型与YouTube-VOSbenchmark中最先进的技术达到了可比较的结果,并且优于以前所有视频对象分割方法,而不是在DAVIS-2017基准测试中使用在线学习。此外,我们的模型比以前的方法实现更快的推理运行时间,在P100GPU上达到44ms /帧。
Multiple object video object segmentation is a challenging task, speciallyfor the zero-shot case, when no object mask is given at the initial frame andthe model has to find the objects to be segmented along the sequence.In ourwork, we propose a Recurrent network for multiple object Video ObjectSegmentation (RVOS) that is fully end-to-end trainable.Our model incorporatesrecurrence on two different domains: (i) the spatial, which allows to discoverthe different object instances within a frame, and (ii) the temporal, whichallows to keep the coherence of the segmented objects along time.We train RVOSfor zero-shot video object segmentation and are the first ones to reportquantitative results for DAVIS-2017 and YouTube-VOS benchmarks.Further, weadapt RVOS for one-shot video object segmentation by using the masks obtainedin previous time steps as inputs to be processed by the recurrent module.Ourmodel reaches comparable results to state-of-the-art techniques in YouTube-VOSbenchmark and outperforms all previous video object segmentation methods notusing online learning in the DAVIS-2017 benchmark.Moreover, our model achievesfaster inference runtimes than previous methods, reaching 44ms/frame on a P100GPU.机器学习论文:RVOS:用于视频对象分割的端到端循环网络(RVOS: End-to-End Recurrent Network for Video Object Segmentation) dTtnYNFIgHA1Ilhg.jpg
URL地址:https://arxiv.org/abs/1903.05612     ----pdf下载地址:https://arxiv.org/pdf/1903.05612    ----机器学习论文:RVOS:用于视频对象分割的端到端循环网络(RVOS: End-to-End Recurrent Network for Video Object Segmentation)
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

kkkgame当前离线
新手上路

查看:236 | 回复:0

快速回复 返回顶部 返回列表