Multi-Scale Feature Selective Matching Network for Object Detection

Numerous deep learning-based object detection methods have achieved excellent performance. However, the performance on small-size and positive negative sample imbalance problems is not satisfactory. We propose a multi-scale feature selective matching network (MFSMNet) to improve of alleviate problems. First, we construct semantic enhancement module (MSEM) compensate for information loss small-sized targets during down-sampling by obtaining richer from features at multiple scales. Then, design anchor (ASM) strategy training dominated samples caused samples, which converts offset values localization branch output in head into scores reduces discarding low-quality anchors. Finally, series quantitative qualitative experiments Microsoft COCO 2017 PASCAL VOC 2007 + 2012 datasets show that our method competitive compared nine other representative methods. MFSMNet runs GeForce RTX 3090.