1 paper
Tomas Sipka, Milan Sulc, Jiri Matas
In many computer vision classification tasks, class priors at test time often differ from priors on the training set. In the case of such prior shift, classifiers must be adapted c…