ICIoU: Improved Loss Based on Complete Intersection Over Union for Bounding Box Regression

An object detector based on convolutional neural network (CNN) has been widely used in the field of computer vision because its simplicity and efficiency. The average accuracy CNN model detection results is greatly affected by loss function. precision localization algorithm function main factor affecting result. Based complete intersection over union (CIoU) function, an improved penalty proposed to improve accuracy. Specifically, more comprehensively considers matching bounding boxes between prediction with ground truth, using proportional relationship aspect ratio from both boxes. Under same two boxes, influence factors box were considered. In this way, strengthened, improved. This called Improved CIoU (ICIoU). Experiments Udacity, PASCAL VOC, MS COCO datasets have demonstrated effectiveness ICIoU improving models one-stage YOLOv4. Compared CIoU, (AP) 0.57% AP75 0.12% AP 0.26% 1.28% 0.06% 0.65% COCO.