Implementation of ['YOLO-OB: An improved anchor-free real-time multi-scale colon polyp detector in colonoscopy']

This repo is the implementation of ['YOLO-OB: An improved anchor-free real-time multi-scale colon polyp detector in colonoscopy'] The deep learning-based object detection algorithm is currently one of the most effective methods for improving polyp detection during colonoscopy. This repository proposes a new model called YOLO-OB based on YOLOv3. It improves network's feature fusion capacity by adopting the Spatial Pyramid Pooling (SPP) and the Bi-directional Feature Pyramid Network (BiFPN) modules. It also employs the ObjectBox detection head, which uses a center-based anchor-free box regression strategy that can detect polyps of different sizes on feature maps of any scale.