computer-vision - Confusion regarding Conditional Random Fields

Question

http://i.imgur.com/dspFhlO.png

I am trying to label objects in am image using Conditional Random Fields. But I am stuck understanding this formula. Can anyone tell me the meaning the terms of the formula and how to calculate them. I am using MS-COCO data set which has labelled images i.e I have segmented images.

Here Z(.)= partition function and P(ci | Sj)= Probability that Sj segment of Image I belongs to class ci and q= no of pairwise spatial relations.

Answer 1

这实际上是c={c1,c2,...,ck}给定片段特征的图像片段标记的条件概率分布S={S1,S2,...,Sk}。p(ci|Si)是将类别标签分配ci给段的概率i，可以使用逻辑回归、神经网络或 SVM 等各种分类器来计算。该术语B表示聚合成对函数，该函数确定每个相邻对{i,j}获取标签的可能性{ci,cj}。这个术语可以通过计算数据集中不同类对的共现统计来实现，在这篇论文中有详细描述： Object Categorization using Co-Occurrence, Location and Appearance