知识卡片
非极大抑制贪心去重复
NMS反复保留最高分框,删除与其IoU超过阈值的候选,直到集合空。实际通常按类别或明确跨类规则处理;高重叠可能来自真实相邻实例,因此阈值过低会误删拥挤对象,不能把重叠直接当同一实体证明。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 目标识别
原书出处 · 1 处
- · 查看原书依据
The nonmaximum suppression algorithm takes as input a set of object bounding boxes and confidences, and outputs a smaller set removing overlapping bounding boxes. It is an iterative algorithm as follows: (1) Take the highest confidence bounding box from the set S and add it to the final set S*. (2) Remove from S the selected bounding box and all the bounding boxes with an
参考来源
- 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第50章《目标识别》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)/_epub-src/OPS/xhtml/c050b.xhtml)
- 结论依据:摘录中关于非极大抑制贪心去重复的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。
- 原始内容:The nonmaximum suppression algorithm takes as input a set of object bounding boxes and confidences, and outputs a smaller set removing overlapping bounding boxes. It is an iterative algorithm as follows: (1) Take the highest confidence bounding box from the set S and add it to the final set S*. (2) Remove from S the selected bounding box and all the bounding boxes with an