Video Object Counting Dataset

摘要

Object counting is an important field of research in computer vision. Unlike other areas such as object recognition, there is serious lack of standard object counting datasets, especially videos, for which the object counting algorithms can be compared. This paper addresses such need by bringing together videos with various challenges. It also gives the object count ground truth that was created using a hybrid human-computer approach to maximise accuracy. The dataset is available freely and is structured in an intuitive way to allow for addition of more object classes.