Efficient Inference in Fully Connected CRFs with Gaussian Edge Potentials

摘要上：

1.CRF的传统方法defined over pixels or image regions，author defined on the complete set of pixels in an image.

2.但是这样的定义making traditional inference algorithms impractical,所以作者主要贡献就是一个高效的估计推理算法。

该论文代码见1.http://www.philkr.net/code/

Krähenbühl P, Koltun V. Efficient inference in fully connected crfs with gaussian edge potentials[C]//Advances in neural information processing systems. 2011: 109-117.

目录：

1.介绍

2.The Fully Connected CRF Model 全连接CRFs模型

3. Efficient Inference in Fully Connected CRFs

   3.1 Mean Field Approximation

   3.2 Efficient Message Passing Using High-Dimensional Filtering

4 Learning

5 Implementation

6 Evaluation

We evaluate the presented algorithm on two standard benchmarks for multi-class image segmen-
tation and labeling. The first is the MSRC-21 dataset, which consists of 591 color images of size 320×213 with corresponding ground truth labelings of 21 object classes [19]. The second is the PASCAL VOC 2010 dataset, which contains 1928 color images of size approximately 500×400,with a total of 20 object classes and one background class [3]. 

其中MSRC=Microsoft Research in Cambridge，MSRC数据集来源于

2.《TextonBoost: Joint Appearance, Shape and Context Modeling for Mulit-Class Object Recognition and Segmentation》

Shotton J, Winn J, Rother C, et al. Textonboost: Joint appearance, shape and context modeling for multi-class object recognition and segmentation[C]//European conference on computer vision. Springer, Berlin, Heidelberg, 2006: 1-15.

同时该论文提出来textonboost算法。

注：DenseCRF作者将MSRC数据集中的不准确的gt重新标注了96张。

为什么要提到2?

1.使用了该数据集

2.

因为一元势函数就是在该算法的基础之上所获得的。