CPSS-FAT: A consistent positive sample selection for object detection with full adaptive threshold.
Pattern Recognit.(2023)
摘要
•Discussing limitations of IoU-based and center-point assignment strategies and revealing their inconsistencies in the open world.•Weight IoU score with the normalized center-distance for better alignment of anchor-object named the consistent positive sample selection (CPSS) method.•To keep conformity with CPSS, introduce a consistent location quality estimation method.•According to the occlusion degree of an object, adopt the full adaptive threshold method to find out more objects.•Achieving the best performance overperforming most popularized detectors on MS COCO 2017.
更多查看译文
关键词
Object detection, Label assignment, Location quality estimation, Adaptive threshold
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要