人工智能培训

搜索

人工智能论文:METAANCHOR:学习使用自定义锚点检测对象(MetaAnchor: Learning to Detect Objects with Cust

[复制链接]
trueno 发表于 2018-7-4 09:16:35 | 显示全部楼层 |阅读模式
trueno 2018-7-4 09:16:35 2219 0 显示全部楼层
人工智能论文:METAANCHOR:学习使用自定义锚点检测对象(MetaAnchor: Learning to Detect Objects with Customized Anchors)我们为objectdetection框架提出了一种名为MetaAnchor的新颖灵活的锚机制。与许多以前的检测器模型锚点不同,通过预定义的方式,在MetaAnchor中,锚点函数可以从任意定制的先前框中动态生成。利用权重预测,MetaAnchor能够与大多数基于锚的对象检测系统(如RetinaNet)协同工作。与预定义的anchorscheme相比,我们凭经验发现MetaAnchor对锚点设置和边界框分布更具鲁棒性;此外,它还显示了潜在的转移任务。我们对COCO检测任务的实验表明,MetaAnchor在各种情况下一致地优于同行。
We propose a novel and flexible anchor mechanism named MetaAnchor for objectdetection frameworks.Unlike many previous detectors model anchors via apredefined manner, in MetaAnchor anchor functions could be dynamicallygenerated from the arbitrary customized prior boxes.Taking advantage of weightprediction, MetaAnchor is able to work with most of the anchor-based objectdetection systems such as RetinaNet.Compared with the predefined anchorscheme, we empirically find that MetaAnchor is more robust to anchor settingsand bounding box distributions;in addition, it also shows the potential ontransfer tasks.Our experiment on COCO detection task shows that MetaAnchorconsistently outperforms the counterparts in various scenarios.人工智能论文:METAANCHOR:学习使用自定义锚点检测对象(MetaAnchor: Learning to Detect Objects with Customized Anchors) Xl3lniK288iZ6IH2.jpg
URL地址:https://arxiv.org/abs/1807.00980     ----pdf下载地址:https://arxiv.org/pdf/1807.00980    ----人工智能论文:METAANCHOR:学习使用自定义锚点检测对象(MetaAnchor: Learning to Detect Objects with Customized Anchors)
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

trueno当前离线
新手上路

查看:2219 | 回复:0

快速回复 返回顶部 返回列表