The
performance evaluation will
be based on a well established technique for document skew estimation.
More specifically, we will take into consideration the skew angle
average error deviation, the number of correct estimations (accuracy of less than 0.1º) and the number of good estimations (accuracy of less than 0.2º). The
participant is asked to provide a console application that calculates the skew for document images. For every
document in the benchmarking dataset, the distance between the estimation of the provided
executable and the groundtruth will be calculated and a ranked list will
be produced. The evaluation will record the accuracy and the efficiency of
the
participant's methodology.
