← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8890 · 生成模型:GPT-6

可分离高斯减少核乘法次数

理解等级:中级

轴对齐、独立维度的高斯可按每轴一维卷积实现。宽度K的二维核直接逐点需约K²次贡献,分离后约2K次;图像总成本还乘输出像素数。等价要求离散核和截断可分离且边界协议一致,旋转的各向异性核未必同样成立。

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

      time when applying the convolution of equation (17.9). If the 2D convolution kernel is N × N samples, then a direct convolution of that 2D kernel scales in proportion to N2, since equation (17.9) requires one multiplication per image position per kernel sample. Using the cascade of two 1D kernels, resulting in an equivalent 2D filter of the same size, scales in proportion to 2N.

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第17章《模糊滤波器》(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/c017b.xhtml) - 结论依据:摘录中关于可分离高斯减少核乘法次数的定义、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:time when applying the convolution of equation (17.9). If the 2D convolution kernel is N × N samples, then a direct convolution of that 2D kernel scales in proportion to N2, since equation (17.9) requires one multiplication per image position per kernel sample. Using the cascade of two 1D kernels, resulting in an equivalent 2D filter of the same size, scales in proportion to 2N.