Bridging the Gap Between Anchor-based and Anchor-free Detection via Adaptive Training Sample Selection 通过自适应训练样本选择弥补基于Anchor和无Anchor检测之间的差距.
Generalized Focal Loss: Learning Qualified and Distributed Bounding Boxes for Dense Object Detection GFocal Loss: 为密集目标检测学习合格且分散的边界框.