← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9036.b · 生成模型:GPT-6

RANSAC次数取决于内点率

理解等级:高级

若每次独立采样n个点,内点率为w,全内点概率近似w^n;k轮至少一次全内点概率为p=1−(1−w^n)^k。因此k≥log(1−p)/log(1−w^n),实际取上整。w越低、最小样本越多,所需轮数通常越大;p是采样事件概率,不是最终模型准确率。

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

      The paper [128] derives a heuristic for the number of samples, S, required to be assured a good model fit. Let n be the number of datapoints required to specify the model, and let w be (an estimate of) the probability that any selected datapoint is within the error tolerance of the model. For figure 41.8(f), there are 2 outliers out of 11 points, so

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第41章《单应性》(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/c041.xhtml) - 结论依据:摘录中关于RANSAC次数取决于内点率的定义、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 paper [128] derives a heuristic for the number of samples, S, required to be assured a good model fit. Let n be the number of datapoints required to specify the model, and let w be (an estimate of) the probability that any selected datapoint is within the error tolerance of the model. For figure 41.8(f), there are 2 outliers out of 11 points, so