← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8913 · 生成模型:GPT-6

最近邻须定义坐标与并列

理解等级:中级

最近邻插值取离目标坐标最近的输入样本,结果分片常数。恰好落在两点中间时需要明确并列规则,图像边缘还需指定外推。输入输出网格的对齐方式会决定每个像素复制到哪里,因此仅说“最近邻放大两倍”不足以唯一规定结果。

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 nearest neighbor interpolation we assign to ℓout(a, b) the closest value of the input ℓin: (21.25) where n – 0.5 < a < n + 0.5 and m – 0.5 < b < m + 0.5. Note that here we are treating ℓout(a, b) as a continuous image. Nearest neighbor interpolation results in an image that is piecewise-constant. Figure 21.18 shows the nearest neighbor

    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:摘录中关于最近邻须定义坐标与并列的定义、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 nearest neighbor interpolation we assign to ℓout(a, b) the closest value of the input ℓin: (21.25) where n – 0.5 < a < n + 0.5 and m – 0.5 < b < m + 0.5. Note that here we are treating ℓout(a, b) as a continuous image. Nearest neighbor interpolation results in an image that is piecewise-constant. Figure 21.18 shows the nearest neighbor