For the competition, we use the PASCAL Visual Object Classes (VOC) detection metric [http://cocodataset.org/#detection-eval]. This metric is designed to evaluate visual object detection and recognition models, which helps machine learning researchers to have standard evaluation procedures. After computing the true positives and false negatives number for each category (object class), the average recall is calculated in addition to the average precision computed by the API. In particular we can use the COCO API protocol4 to evaluate the results of the ICDAR2019-ORF competition because of its features which are based on the size of the objects and also the computation of the average recall in addition to the precision.