An Object Detector based on Bi-directional Feature Pyramid Network and Faster R-CNN
Bo Jin, Tao Zhang, Bin Tang, Qin Zhang · 2022
Feature pyramids are a basic component in recognition systems for detecting objects at different scales. The proposal of Feature Pyramid Networks(FPN) has achieved great success in the field of object detection and greatly improved the detection accuracy, but FPN is only a simple top-down single feature fusion.Therefore, in order to build a higher-level feature pyramid with both semantic information and location-spatial information, we propose a top-down and bottom-up bi-directional Feature Pyramid Network(BiFPN). Using BiFPN in a basic Faster R-CNN system, our method can achieve a state-of-the-art mAP of 82.2% on PASCAL VOC dataset, surpassing all improved methods based on the Faster R-CNN system.