人工智能培训

搜索

机器学习论文:引导锚定的区域提案(Region Proposal by Guided Anchoring)

[复制链接]
犀牛 发表于 2019-1-11 10:46:17 | 显示全部楼层 |阅读模式
犀牛 2019-1-11 10:46:17 80 0 显示全部楼层
机器学习论文:引导锚定的区域提案(Region Proposal by Guided Anchoring)区域锚点是现代物体检测技术的基石。现有技术的探测器主要依赖于密集的锚定方案,其中在空间域上均匀地采样具有预定义的一组体积和纵横比。在本文中,我们重新审视了这个基础阶段。我们的研究表明,它可以更加有效和高效地完成。具体来说,我们提出了一种替代方案,称为引导锚定,它利用语义特征来指导锚定。所提出的方法共同预测了可能存在感兴趣对象的中心的位置以及不同位置处的尺度和纵横比。在预测的锚形状之上,我们通过featureadaption模块减轻了特征的不一致性。我们还研究使用高质量的建议来提高检测性能。锚定方案可以是无缝集成的toproposal方法和检测器。通过Guided Anchoring,我们在MS COCO上的召回率提高了9.1美元或更高,其锚点比RPN基线少90%。%。 Wealso采用快速R-CNN中的引导锚定,更快的R-CNN和RetinaNet,分别将检测mAP提高$ 2.2 \%$,$ 2.7 \%$和$ 1.2 \%$。
Region anchors are the cornerstone of modern object detection techniques.State-of-the-art detectors mostly rely on a dense anchoring scheme, whereanchors are sampled uniformly over the spatial domain with a predefined set ofscales and aspect ratios.In this paper, we revisit this foundational stage.Our study shows that it can be done much more effectively and efficiently.Specifically, we present an alternative scheme, named Guided Anchoring, whichleverages semantic features to guide the anchoring.The proposed method jointlypredicts the locations where the center of objects of interest are likely toexist as well as the scales and aspect ratios at different locations.On top ofpredicted anchor shapes, we mitigate the feature inconsistency with a featureadaption module.We also study the use of high-quality proposals to improvedetection performance.The anchoring scheme can be seamlessly integrated toproposal methods and detectors.With Guided Anchoring, we achieve $9.1\%$higher recall on MS COCO with $90\%$ fewer anchors than the RPN baseline.Wealso adopt Guided Anchoring in Fast R-CNN, Faster R-CNN and RetinaNet,respectively improving the detection mAP by $2.2\%$, $2.7\%$ and $1.2\%$.机器学习论文:引导锚定的区域提案(Region Proposal by Guided Anchoring) WVTiCM8YIToyPTtn.jpg
URL地址:https://arxiv.org/abs/1901.03278     ----pdf下载地址:https://arxiv.org/pdf/1901.03278    ----机器学习论文:引导锚定的区域提案(Region Proposal by Guided Anchoring)
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

犀牛当前离线
新手上路

查看:80 | 回复:0

快速回复 返回顶部 返回列表