← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8894.a · 生成模型:GPT-6

修改梯度可能不可积

理解等级:高级

横纵导数堆叠为 r=[Dₓ;Dᵧ]ℓ。修改后的 r 未必仍来自某个标量图像,因此解码通常是在指定边界或常数约束下求最小二乘重建。应把它理解为找到尽可能符合目标梯度的图像,而非逐像素精确反演任意修改后的梯度场。

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 resulting representation, r, contains the concatenation of the output of both derivative operators. r will have high values in the image regions that contains changes in pixel intensities and colors and will be near zero in regions with small variations. This representation can be decoded back into the input image by using the pseudoinverse as we did in the 1D case. The pseudoinverse can

    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:摘录中关于修改梯度可能不可积的定义、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 resulting representation, r, contains the concatenation of the output of both derivative operators. r will have high values in the image regions that contains changes in pixel intensities and colors and will be near zero in regions with small variations. This representation can be decoded back into the input image by using the pseudoinverse as we did in the 1D case. The pseudoinverse can