LocNet: Improving Localization Accuracy for Object Detection
来源:互联网 发布:wpf编程宝典2013 pdf 编辑:程序博客网 时间:2024/05/16 08:22
这篇论文主要目的是提升检测框与目标的吻合度,特别是当IOU比较大时。之前主要使用bbox回归的方法,作者通过给搜索区域的每列或每行,或在目标bbox内分配概率解决,如下图所示:
检测方法步骤:1.给定候选框,分配置信度;2.给定候选框,放大得到搜索区域,迭代得到新的更接近目标的候选框,算法流程如下:
给定搜索区域R,划分成M个水平区域和竖直区域,返回每个区域的条件概率,考虑了In-Out概率和边界概率两个条件概率,主要是这两个信息对检测结果更有用。作者构建的检测模型即LocNet如下图所示:
检测结果:结果主要是在IOU>0.65时与bbox回归相比,与实际目标的吻合度更高,这个方法对于需要精确目标位置的应用应该有帮助,matlab代码可以试试:https://github.com/gidariss/LocNet。
0 0
- LocNet: Improving Localization Accuracy for Object Detection
- LocNet: Improving Localization Accuracy for Object Detection
- 论文研读--LocNet: Improving Localization Accuracy for Object Detection
- LocNet: Improving Localization Accuracy forObject Detection
- LocNet:Improving LocalizationAccuracy for Object Detection 随笔
- Localization and Object Detection
- 20170324#cs231n#9.ConvNets for spatial localization & Object detection
- hsc for object detection
- DenseBox: Unifying Landmark Localization with End to End Object Detection
- 目标检测--Improving Object Detection With One Line of Code
- Improving Object Detection With One Line of Code
- Improving Object Detection With One Line of Code
- 目标检测:Improving Object Detection With One Line of Code
- Object Detection: To Be Higher Accuracy and Faster
- 论文笔记:Is object localization for free?
- [深度学习论文笔记][Object Localization] OverFeat: Integrated Recognition, Localization and Detection using C
- Regionlets for Generic Object Detection
- regionlets for generic object detection
- Disruptor原理
- Java Web学习(22): 阶段小项目实现商品浏览记录
- 色达
- 学String类 有感
- 深入理解RxJava的Side Effect Methods
- LocNet: Improving Localization Accuracy for Object Detection
- 实时更新线上App:JSPatch
- 一些不错的文档
- Error:C:\Users\lqm\.gradle\caches\2.10\scripts\ijinit34_7wu3ex74z3a8e98fc8d35fuid\cp_init\cache.prop
- Meta http-equiv属性值X-UA-Compatible
- Android实现沉浸式通知栏通知栏背景颜色跟随app导航栏背景颜色而改变
- 图片拍照图片 处理工具 旋转 存贮等等
- 转载 机器学习--正则化理解
- 24. Swap Nodes in Pairs