Anchor Box Parameters and Bounding Box Overlap Ratios for the Faster R-CNN Detector in Detecting a Single Object by the Masking Background

2018
Vadim Romanuke

Anchor box parameters and bounding box overlap ratios are studied in order to set them appropriately for the Faster R-CNN detector. The benchmark detection is based on monochrome images whose background may mask a small dark object. Three object detection tasks are generated, where every image either contains a small black square/rectangle or does not contain the object, representing thus class “background”. If this class is represented, the ratios are recommended to be tried at 0.7. The ratio for positive training samples is to be tried at a less value, but greater than 0.4, for the task every image of which contains an object. The minimum anchor box size is better to be tried at a lesser value from a range of object sizes. The anchor box pyramid scale factor and the number of levels are better to try at 2 and 8, respectively. Subsequently, these parameters may be corrected as their influence is fuzzier than that of the ratios.

Atslēgas vārdi
Anchor box, bounding box overlap ratio, object detection, R-CNN.
DOI
10.7250/itms-2018-0002
Hipersaite
https://itms-journals.rtu.lv/article/view/itms-2018-0002

Romanuke, V. Anchor Box Parameters and Bounding Box Overlap Ratios for the Faster R-CNN Detector in Detecting a Single Object by the Masking Background. Information Technology and Management Science, 2018, Vol. 21, No. 1, 15.-23.lpp. ISSN 2255-9086. e-ISSN 2255-9094. Pieejams: doi:10.7250/itms-2018-0002

Publikācijas valoda
English (en)