A Statistical Confidence Measure for Optical Flows-论文阅读
来源:互联网 发布:管家婆软件怎么安装 编辑:程序博客网 时间:2024/05/22 12:55
1. 摘要
提到confidence measure有四点好处:1. 可以检测到不可靠的光流; 2. 利用in-painting技术补全剔除的不可靠光流区域;3. 将可靠性度量和变分法结合,提高光流的性能 4. 利用置信度和低复杂度的光流技术替换高复杂度的光流技术。
2. 相关文章综述
利用intrinsic dimension的原理进行confidence measure...
一些论文利用输入图像本身的特性,一些论文的confidence算法和optical flow本身直接相关。
3. Natural Motion Statistical
To compute the statistical model, the empirical mean m and covarianceC are computed from the training data set containing the n n T flow patches, which
are vectorized in lexicographical order.
从training data中计算出均值m和协方差C。
均值m是2*n*n*T维度。可以这么理解:从测试序列中选取若干个(10000/5000个随意)patch (N),每个patch的维度是(mvx0, mvx1, .... mvxn*n*T, mvy0, mvy1, .... mvyn*n*T);
对N个patch的对应维度进行求均值,可以得到各个维度下的均值,同理也可以得到对应的协方差矩阵。
其实可以这么理解,这里的随机变量有2*n*n*T个,patch中的每一个像素位置的mvx和mvy就是对应一个随机变量。
4. Hypothesis Testing
假设条件概率密度是二维的正态分布 the conditional pdf is a two dimensional normal distribution.
最终的Confidence函数如下所示:
其中alpha为置信度,自己可配置。
G函数的横坐标是Mahalanobis distance(dm)的,纵坐标是0到1之间,即对训练样本计算出dm后的分布曲线。
其中G-1为quantile function,如下所示:G为training数据的分布特性
dM为the squared Mahalanobis distance
Mahalanobis distance wiki: https://en.wikipedia.org/wiki/Mahalanobis_distance
- A Statistical Confidence Measure for Optical Flows-论文阅读
- 论文阅读:Combining volumetric dental CT and optical scan data for teeth modeling
- [人眼检测] high confidence visual recognition of persons by a test of statistical independence
- 阅读笔记Surflet-Pair-Relation Histograms: A Statistical 3D-Shape Representation for Rapid Classification
- [文献阅读] A Statistical MT Tutorial Workbook
- 【论文阅读笔记】Deepr: A Convolutional Net for Medical Records
- XSL Transformation for mediation flows
- 软硬件划分论文阅读系列 2 -- A Configurable Logic Architecture for Dynamic Hardware Software Partitioning
- 论文阅读——译文:Chord: A Scalable Peer-to-peer Lookup Service for Internet Applications
- 论文阅读笔记 - Mesos: A Platform for Fine-Grained ResourceSharing in the Data Center
- 论文阅读:A Critical Review of Recurrent Neural Networks for Sequence Learning
- 论文阅读:A Discriminative Feature Learning Approach for Deep Face Recognition
- 论文阅读:CVPR 2015 FaceNet: A Unified Embedding for Face Recognition and Clustering
- A Simple Deep and Effective Neural Network for Semantic Role Labelling 论文阅读
- 论文阅读:A Bottom-Up Approach for Automatic Pancreas Segmentation in Abdominal CT Scans
- [NLP论文阅读]A simple but tough-to-beat baseline for sentence embedding
- 【论文阅读】Sequential Matching Network: A New Architecture for Multi-turn Response Selection in Retrieval
- 经典论文阅读——A Convolutional Neural Network Cascade for Face Detection
- c指针和malloc的运用注意
- android:windowSoftInputMode属性详解
- 文件系统超级块
- js控制文本字数,并能点击显示和隐藏
- 关于的servlet的单例模式解释
- A Statistical Confidence Measure for Optical Flows-论文阅读
- oracle数据库小结
- 基于spring mvc注解项目 启动时初始化数据
- android:screenOrientation
- nginx反向代理TCP,取RTMP流
- 《科学》封面重磅论文:人工智能终于能像人类一样学习
- srcollview 嵌套ListView ListView 再嵌套gridview 焦点滑动问题
- javascript——textarea自动伸缩问题
- 设置tabBarItem的图片渲染