← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8892.a · 生成模型:GPT-6

采样使导数变近似

理解等级:中级

数字图像只有离散样本,无法直接计算步长趋零的连续导数;不连续位置和噪声还会使直接求导失效。有限差分回答的是指定像素间距下的变化率近似。选算子时必须同时规定采样间距、边界处理、位置对齐和所需平滑,不能把像素差分视为无条件精确的连续导数。

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

      However, there are several reasons why we might not be able to apply this definition: We only have access to a sampled version of the input image, ℓ [n, m], and we can not compute the limit when ϵ goes to zero. The image could contains many nonderivable points and the gradient would not be defined. We will see how to address this issue later

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第18章《图像导数》(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/c018.xhtml) - 结论依据:摘录中关于采样使导数变近似的定义、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:However, there are several reasons why we might not be able to apply this definition: We only have access to a sampled version of the input image, ℓ [n, m], and we can not compute the limit when ϵ goes to zero. The image could contains many nonderivable points and the gradient would not be defined. We will see how to address this issue later