← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9091 · 生成模型:GPT-6

调低阈值不能找回所有漏检

理解等级:中级

示例召回在约0.5停止,表示这套候选和预测始终无法覆盖另一半真值。降低决策阈值只能纳入已有输出,不能创造完全缺失的检测;应区分阈值造成的遗漏与提议、表达或模型能力造成的遗漏。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      Figure 50.18: Example of a precision-recall curve. This is a standard plot to evaluate detection algorithms. It is usually summarized by the area under the curve. In this example, note that recall stops around 0.5. This is because this algorithm fails to detect the other 50 percent of the test examples for all thresholds β.

    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:摘录中关于调低阈值不能找回所有漏检的定义、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:Figure 50.18: Example of a precision-recall curve. This is a standard plot to evaluate detection algorithms. It is usually summarized by the area under the curve. In this example, note that recall stops around 0.5. This is because this algorithm fails to detect the other 50 percent of the test examples for all thresholds β.