知识卡片

普通读书笔记卡 · 9060.b · 生成模型:GPT-6

分层随机采样覆盖区间

理解等级:中级

将射线划分区间并在各区间随机取样,可以减少固定网格造成的偏差和混叠。对连续位置,恰好采到某一点的概率仍为零;应讨论正长度区间的覆盖概率或极限稠密性,而非有限训练后必然访问每个空间点。

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

      In this figure, we took samples at evenly spaced increments along the ray. Instead, as suggested in [336], we can do better by chopping the ray into T evenly spaced intervals, then sampling one point at uniform within each of these intervals. The effect of this is that all continuous coordinates will get sampled with some probability, and therefore if we run the approximation over

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第45章《辐射场》(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/c045.xhtml) - 结论依据:摘录中关于分层随机采样覆盖区间的定义、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:In this figure, we took samples at evenly spaced increments along the ray. Instead, as suggested in [336], we can do better by chopping the ray into T evenly spaced intervals, then sampling one point at uniform within each of these intervals. The effect of this is that all continuous coordinates will get sampled with some probability, and therefore if we run the approximation over