Platform for Teaching Detection Classifiers

Object Detection, AdaBoost, WaldBoost, Machine Learning, Teaching

Scanning using very fast classifiers is a standard and successful approach to object detection in images. This approach became popular after Viola and Jones introduced their frontal face detector in 2001 which was able to reliably detect faces in unconstrained condition and in real-time. Although detecting objects by these classifiers is taught in many computer vision courses, it is not possible for the students to experiment with the methods because the existing implementations require significant initial effort to be able to train and test a classifier and to interpret the detection results in an intuitive way. In this paper, we describe a web-based application which allows experimenting with detection classifiers with minimal initial effort. The application has an intuitive user interface which allows for simple configuration of experiments and management of results. It also provides pre-prepared experiments and datasets in order to further reduce the initial effort. The application is publicly available so anyone can experiment with the detection classifiers and also train detectors on their own data. The application has a potential to become a useful teaching tool for lecturers of computer vision courses and for other interested people.