← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8914.b · 生成模型:GPT-6

重复边界影响最后样本

理解等级:中级

书中的二倍上采样矩阵交替输出原样本与相邻平均,最后一个位置因重复边界而复制末样本。这个边缘结果来自外推规则,不是所有双线性放大的唯一输出。对照库函数时必须先统一坐标中心、对齐方式、输出长度与padding。

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 upsampling operator is also a linear operation, therefore we can also write it in matrix form. For instance, for a 1D signal of length N = 4 samples, if we upsample it by a factor k = 2, then, using repeat padding for handling the boundary and the bilinear interpolation, we can write the upsampling operator as a matrix:

    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:摘录中关于重复边界影响最后样本的定义、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 upsampling operator is also a linear operation, therefore we can also write it in matrix form. For instance, for a 1D signal of length N = 4 samples, if we upsample it by a factor k = 2, then, using repeat padding for handling the boundary and the bilinear interpolation, we can write the upsampling operator as a matrix: