← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8912.a · 生成模型:GPT-6

下采样矩阵包含边界

理解等级:中级

用 [1,2,1]/4 平滑后抽取,内部输出是邻域加权平均。若首边界重复首像素,首行会出现3/4、1/4权重,和普通内部模板不同。矩阵表达揭示边界约定也是算子的一部分;换一种padding就改变整个有限图像的变换。

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

      As the convolution and downsampling are linear operators, sometimes it will be useful to write it as a matrix. For instance, for a 1D signal of length N = 8 samples, if we downsample it by a factor k = 2, then, using repeat padding for handling the boundary, and the binomial kernel as an anti-aliasing filter, we can write the downsampling operator as a

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第21章《图像下采样与上采样》(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/c021.xhtml) - 结论依据:摘录中关于下采样矩阵包含边界的定义、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:As the convolution and downsampling are linear operators, sometimes it will be useful to write it as a matrix. For instance, for a 1D signal of length N = 8 samples, if we downsample it by a factor k = 2, then, using repeat padding for handling the boundary, and the binomial kernel as an anti-aliasing filter, we can write the downsampling operator as a