Multi-Level Refinement Feature Pyramid Network for Scale Imbalance Object Detection

Object detection becomes a challenge due to diversity of object scales. In general, modern object detectors use feature pyramid to learn multi-scale representation for better results. However, current versions of feature pyramid are insufficient to handle scale imbalance, as it is inefficient to integrate semantic information across different scales. Here, we reformulate feature pyramid construction as a feature reconfiguration process. We propose a detection network, Multi-level Refinement Feature pyramid Network, to combine high-level features (i.e., semantic information), middle-level feature and low-level feature (i.e., boundary information), in a highly-nonlinear yet efficient manner. A novel contextual features module is proposed, which consists of global attention and local reconfigurations. It efficiently gathers task-oriented contextual features across different scales and spatial locations (i.e., lightweight local reconfiguration and global attention). To evaluate significance of proposed model, we designed and trained end-to-end single stage detector called MRFDet by assimilating it into Single Shot Detector (SSD), and it achieved better detection performance compared to most recent single-stage objects detectors. MRFDet achieves an AP of 45.2 with MS-COCO and an improvement in mAP of 4.5% with VOC.