Semantic Segmentation of Small Objects and Modeling of Uncertainty in Urban Remote Sensing Images

评级

C

简介

cvpr workshop 2016, Michael Kampffmeyer， 引用量308，这个和small object 不太一样的是，coco里面每个类都可能有大有小，seg里面类别比如car这种就一定很小，这个文章就是直接通过改变类别频率，不如直接说class of small area

创新

median frequency reweight

We study the potential of the cross-entropy loss function weighted with median frequency balancing, a loss function that was proposed by Eigen and Fergus [9] in the last few months, to improve segmentation accuracy for small classes in urban remote sensing,

感觉没什么用，不一定比直接权重加大效果好

网络没什么创新

发现改变阈值可以提高acc

这不废话嘛