人工智能培训

搜索

论文代码开源:30FPS的R-FCN-3000:解耦检测和分类(R-FCN-3000 at 30fps: Decoupling Detection and Cl

[复制链接]
admin 发表于 2018-6-24 05:46:27 | 显示全部楼层 |阅读模式
admin 2018-6-24 05:46:27 1516 0 显示全部楼层
人工智能论文代码开源:30FPS的R-FCN-3000:解耦检测和分类(R-FCN-3000 at 30fps: Decoupling Detection and Classification)请注意该人工智能论文代码开源在github,大部分是python写的,框架可能是tensorflow或者pytorch。我们提出了R-FCN-3000,一种大型实时对象检测器,其中对象检测和分类是解耦的。为了获得RoI的检测结果,我们将对象分数乘以细粒度分类得分。我们的方法是修改R-FCN架构,其中位置敏感滤波器跨不同对象类共享以执行本地化。对于细粒度分类,不需要这些位置敏感滤波器。 R-FCN-3000在ImageNet检测数据集上获得了34.9%的mAP,并且在每秒处理30幅图像的同时,胜过YOLO-900018%。我们还表明,由R-FCN-3000学习的对象被推广到新的类中,并且性能随着训练对象类的数量增加而增加 - 支持可以学习通用对象检测器的假设。代码将可用。
We present R-FCN-3000, a large-scale real-time object detector in whichobjectness detection and classification are decoupled.To obtain the detectionscore for an RoI, we multiply the objectness score with the fine-grainedclassification score.Our approach is a modification of the R-FCN architecturein which position-sensitive filters are shared across different object classesfor performing localization.For fine-grained classification, theseposition-sensitive filters are not needed.R-FCN-3000 obtains an mAP of 34.9%on the ImageNet detection dataset and outperforms YOLO-9000 by 18% whileprocessing 30 images per second.We also show that the objectness learned byR-FCN-3000 generalizes to novel classes and the performance increases with thenumber of training object classes - supporting the hypothesis that it ispossible to learn a universal objectness detector.Code will be made available.论文代码开源:30FPS的R-FCN-3000:解耦检测和分类(R-FCN-3000 at 30fps: Decoupling Detection and Classification) kh4z8kK43hTKXzVk.jpg
URL地址:https://arxiv.org/abs/1712.01802v1     ----pdf下载地址:https://arxiv.org/pdf/1712.01802v1    ----         ----github下载地址:https://github.com/MahyarNajibi/SNIPER    ----    论文代码开源:30FPS的R-FCN-3000:解耦检测和分类(R-FCN-3000 at 30fps: Decoupling Detection and Classification)请注意该人工智能论文代码开源在github,大部分是python写的,框架可能是tensorflow或者pytorch,keras,至于具体是哪一个没有完全测试。
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

admin当前离线
管理员

查看:1516 | 回复:0

快速回复 返回顶部 返回列表