Improving Loss Function for Polyp Detection Problem

The utilization of automatic polyp detection during endoscopy procedures has been shown to be highly advantageous by decreasing the rate of missed detection by endoscopists. In this paper, we propose a new loss function for training an object detector based on the EfficientDet architecture to detect polyp areas in endoscopic images. The proposed loss combines the features of the Focal loss and DIoU (Distance Intersection over Union) loss named as Focal-DIoU. In addition, we have also carried out some experiments to evaluate the proposed loss function. The experimental results show that our proposed model achieves higher accuracy than previous works on two public datasets. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023.